Dean Kamen unveiled the Segway on December 3, 2001, on Good Morning America, where Diane Sawyer said on air, “I’m tempted to say, that’s it?” The coverage that followed was about the machine: the gyroscopes, the rider who leans forward to go forward. What nobody could say was what it was. A vehicle, a scooter, a toy, a piece of medical equipment. The Segway sold for $4,000 to $5,500, which lands in the gap between two budgets: too much to buy the way you buy a toy, too little to be worth the procurement process that buys municipal vehicles. A buyer who can’t say what kind of thing you are has no way to decide whether she wants one.

The buyer picks your shelf with or without you

Al Ries and Jack Trout turned the problem into a law in The 22 Immutable Laws of Marketing: a company that can’t be first in a category should set up a new category it can be first in. Their concern was memory, which brand comes to mind when the category comes up. The shelf does more than that. It settles which budget pays for you, which title has the authority to sign, and what a fair price looks like, and it settles all three before anyone evaluates a feature.

A shopper who picks up an unfamiliar bottle in a grocery store decides what it is before she reads the label. Soda, juice, supplement, medicine. A CIO who meets a new technology does the same thing. Is it a CRM, a SIEM, an ERP, or something she has no shelf for yet? She’s trying to spend as little effort on the decision as she can, so she’ll place you somewhere whether the fit is good or not. When your category is undefined or keeps moving because you keep rewriting the homepage, no shelf fits.

The distance between two shelves is the distance between two budgets. A product read as a developer tool gets bought on a team’s discretionary line by a director who can sign alone. The same product read as infrastructure goes to a capital review, waits for the annual cycle, and needs a sign-off from the CFO’s office. Nothing about the software changed. What changed is the price it can carry, the number of weeks it takes to close, and the number of people who have to be convinced.

I’ve argued that a buyer who hasn’t been handed a way to read a new thing will supply one herself, reaching for whatever costs her the least thought. An undefined category makes the cheapest available thought I don’t know what this is.

You have two ways out. Pick one deliberately, before launch, rather than finding out at the end of a bad year which one you’ve been half-making.

Borrow a shelf

The first way is to anchor inside a large existing category and take a niche within it. “An AI-driven text-generation engine” gives a buyer nothing to hold on to. “A CRM for freelancers” lands instantly because buyers already know roughly what a CRM costs, which budget it comes out of, and who signs for it. You inherit the parent category’s stability and spend your scarce time explaining your niche instead of on teaching the market what kind of thing you are.

What you inherit along with the shelf is its ceiling. The parent category carries a price the market has already settled on, and charging above it means arguing against an anchor you picked for yourself. You also get graded on the incumbent’s feature list, because a buyer who has filed you under CRM will evaluate you as a CRM, including the twenty things you deliberately didn’t build. Borrowing is the safer path, and it’s the right one when the parent category is mature and your real innovation is the segment rather than the kind of thing.

Salesforce is usually told as a category-creation story, but the more interesting part is what it left alone. At the turn of the millennium sales leaders already knew what CRM was, what Siebel charged for it, and who owned the budget line, and Salesforce asked them to relearn none of it. What it renamed was the delivery model, with the word “software” crossed out in the logo. The familiar shelf carried the price anchor and the buyer along with it, so the whole argument narrowed to how the thing got installed.

I’ve borrowed a shelf myself. When we called Mesosphere’s product a Data Center Operating System, we weren’t asking anyone to learn a new kind of thing. Everyone already knew what an operating system was and knew you couldn’t run a computer without one, and we took fifty years of accumulated understanding for free. We invented the name and borrowed the shelf.

Build a shelf

The second is to build the product and its category together. Higher risk, higher ceiling, and more work than it sounds, because a category isn’t something you can ship. It exists only once other people are using it: when buyers have words for the space and when a company has a line in its budget to spend against. It gets built alongside the product, by you, and it never shows up in a release note. Three jobs come with it, and none of them is engineering.

Standardize the vocabulary so that the market has words for the space when you aren’t in the room. Define what winning looks like, because if you don’t set the measure the incumbents will apply theirs, and theirs was built to make their product win. And make the problem legible before you sell the answer. “Data gravity” is a problem; “container orchestration” is a solution. Categories are built on problems, because a buyer has to agree the problem exists before she’ll accept that a class of product addresses it.

Red Bull is the cleanest example outside software. Dietrich Mateschitz launched it in Austria in 1987 into a European market that had soft drinks and nothing else. The drink itself was not new: Red Bull is a reformulation of Krating Daeng, on sale in Thailand since 1976. What Mateschitz built from scratch was the shelf, and most of that work was vocabulary rather than product: what an energy drink is for, when you drink one, what it’s reasonable to pay. It took years, and the shelf they built now holds every competitor they have.

The Segway is the same path attempted with none of the three jobs done. Kamen built the product and expected the category to arrive on its own, and what vocabulary work happened went to legislatures rather than to buyers. The company had a legal class within three months of the unveiling and never got a budget line. It sold roughly 23,500 units in the U.S. between March 2002 and September 2006. Building a category and building a product are separate jobs, and finishing one of them earns no partial credit on the other.

Your early customers will hide the problem from you

Visionaries and early adopters buy interesting technology on its merits, and they’re happy to be the only company doing something. When you can’t tell them what kind of thing you are, they’ll work it out for themselves and enjoy the puzzle. They’ll take your call, sift through the architecture, and sign. Geoffrey Moore named the wall in Crossing the Chasm in 1991: the early majority buys known solutions to known problems in categories that already exist from vendors their peers can name. Your early wins are real revenue and they’re also camouflage because they came from the one group whose buying behavior doesn’t test the thing that’s broken.

You hit the wall when you run out of visionaries and start selling to the early majority. The damage shows up late in the sales process, far from the category decision that caused it. Cycles get longer. Deals need three more meetings. A champion who loves the product stops returning calls after he takes it to his boss, and the loss reason gets written down as “no decision” or “budget.” What happened is that a buyer who can’t categorize you can’t defend you internally, so ambiguity converts into perceived risk somewhere, usually in a room you weren’t in.

The work nobody schedules

Category work needs investment and stamina because the number that would justify it arrives after the budget cycle that has to fund it. A roadmap review happens every two weeks with a list you can check off. Evangelism has no weekly number, produces nothing demoable, and takes quarters to show up anywhere a board can see it.

Category work is also cheap while it’s still on paper and expensive afterward. Before launch, settling what kind of company you are costs an argument among the founders. After two years with no shelf, it costs a repositioning, with a price change, a sales team retrained off a pitch they’ve internalized, and a customer base that has to be told the thing they bought is now something else.

You can run a test this week without anyone’s permission. Call three customers who’ve been live for a quarter and ask each one how she described you to her boss when she asked for the money. You’re listening for one sentence, without your product name in it, naming a shelf her boss already had. If you get three different sentences, or a sentence that takes a paragraph, you’re already paying for it in cycle time and in losses booked as “no decision,” and the bill goes up every quarter you leave it alone.

Subscribe now

By the time Iridium’s 66 satellites were in orbit, much of their intended market had disappeared. During the company’s eleven-year build, terrestrial carriers extended cellular coverage across the cities where its customers traveled. The executive Iridium had designed the service for could now make the call from Frankfurt or Jakarta with an ordinary mobile phone. Iridium asked that customer to buy a $3,000 satellite handset and pay $3 to $8 a minute.

Iridium spent more than $5 billion building a system that worked. Service opened in November 1998. Nine months later, with roughly 20,000 subscribers against the 52,000 required by its loan covenants, the company filed for Chapter 11. A bankruptcy court later approved its sale for $25 million.

Timing was embedded in Iridium’s business thesis. The company had to reach the places its customers traveled before regular cell phones closed the window. That window was narrowing and then closed while the rockets were still going up.

Every company betting on a market shift carries a similar assumption. A cost must fall, a rule must change, a platform must spread, or buyers must become ready. Calling yourself early explains nothing. You need to name the change, track it, and fund the wait.

Name the Claim

A plan that depends on the market changing should say exactly what has to change. “AI is changing everything” gives you nothing to track. A useful timing claim names evidence you can watch: a product becomes cheap enough for widespread use, a regulation takes effect, a platform adds a capability you need, or customers begin buying in measurable numbers.

Put four things in the plan: the change you expect, the date by which it must happen, the evidence that would invalidate the schedule, and the cash required to survive the wait. Track any technology that could remove the customer problem before your market develops. When the expected evidence fails to appear, change the plan.

Companies often budget market education as marketing, which understates the capital required to wait for buyers. If customers won’t be ready for two years, you need enough cash to cover those two years without assuming revenue from the new market. A product or service that solves a current problem can provide that revenue while the larger market catches up.

Waiting for certainty creates a different timing problem. Once the evidence makes the opportunity feel safe, your competitors can see it too. Companies with more money will enter quickly. You have to act while customers are becoming ready and the market still has room for you to claim a clear position.

I’ve Paid for the Wait

At Vapor IO, we bet that latency and bandwidth economics would make distributed compute near the user inevitable. We built the infrastructure, helped define edge computing, and won some of the world’s largest infrastructure providers as customers and partners. Much of the work was still market education. We had to explain why computing needed to move beyond centralized cloud regions before buyers could decide where edge infrastructure belonged in their plans.

We raised a lot of money. Nearly $100 million. That capital extended the time available for market education while customer budgets and production deployments continued on their own schedules. The larger window never opened wide enough, soon enough, for us.

That experience changed how I think about market timing. A technical forecast can tell you where costs and capabilities are headed. The commercial forecast must account for how long customers need; every month between technical possibility and commercial demand has to be financed.

Early Companies Finance the Market

A company that arrives before the shift spends its money on market education, often to the benefit of competitors. You write the explainer that defines the problem. You fund the conference track. You brief the analysts twice a year until they stop asking what the words mean. You make the same argument in three hundred sales calls, and slowly the category becomes legible to buyers who had no vocabulary for it. Then the money runs out. Two years later a better-timed entrant walks into a room where the buyer already knows what the thing is, already believes She needs it, and wants to talk about implementation and price. You paid for that room. The competitor books the revenue in it.

The accounting offers no mercy. A company that’s early for the right reasons reports the same missed plan as one built on a bad premise. The board and investors see the same revenue shortfall. By the time customers catch up, the original company is often gone, and the competitor that survived long enough collects the proof and the revenue.

General Magic is the cleanest case I know. Marc Porat, Bill Atkinson, and Andy Hertzfeld spun it out of Apple in May 1990 to build the handheld personal communicator. Magic Cap was the operating system. Telescript was a language for software agents that could travel across a network and run errands for you, something that people are still trying to ship. Sony’s Magic Link went on sale in September 1994 at about $995. Motorola’s Envoy followed that December at $1,500.

General Magic’s 1996 annual report names the problem. Telescript, the filing says, “was designed for a proprietary telecommunications network environment,” and the market for personal communicators “has been slow to develop.” The Web showed up and took the electronic marketplace General Magic had spent six years describing. AT&T shut down PersonaLink, the only commercial Telescript service, in August 1996.

What General Magic paid for became other people’s products. Tony Fadell left for Apple and ran the iPod, then the iPhone hardware. Andy Rubin, lead engineer on the Envoy, went on to found Android. Pierre Omidyar launched AuctionWeb, which became eBay, while still on General Magic’s payroll. The company discontinued operations in September 2002 and told shareholders to expect nothing. The handheld communicator it had described in 1990 shipped as the iPhone five years later.

Track the Curve That Can Erase Demand

Iridium’s engineers had identified a real constraint. Geostationary satellites orbited roughly 22,000 miles above Earth, which added delays to calls and required powerful radios with large antennas. Motorola’s proposed low-Earth-orbit constellation would make faster connections and smaller handsets possible. The engineering plan tracked that curve closely from 1987 until service opened eleven years later.

The business depended on another curve moving slowly. Terrestrial cellular coverage determined how many travelers would still need a satellite phone once Iridium was ready. Sydney Finkelstein and Shade Sanford traced the collapse in “Learning from Corporate Mistakes: The Rise and Fall of Iridium”. Cellular networks spread much faster than Iridium’s plan assumed.

John Richardson, who took over as interim chief executive, attributed the failure thusly: “First we created a marvelous technological achievement. Then we asked how to make money on it.” Under new owners, Iridium found a smaller market and remains in business. Its original plan failed because the window closed.

What You Can’t Know From Inside

Timing is much clearer in hindsight. Looking back, people can point to the year a market took off. While you’re waiting, you know only that the change seems real and customers haven’t responded yet. You can’t tell whether they’ll be ready in eighteen months or five years. A company can survive the first delay and run out of money during the second.

That uncertainty belongs in the cash plan and every board discussion about growth. At each review, decide how much more cash the evidence justifies. A missed date should change the plan; a competing technology that removes the need should end the original bet.

Subscribe now

A go-to-market plan usually arrives secondhand. A founder names three companies she admires, somebody reconstructs how those companies launched, and the plan gets assembled out of the parts that worked: the content cadence, the pricing page, the conference booth, the analyst briefings. The tactics are described accurately. What travels with them, unannounced, is a set of assumptions about the market the original company was standing in, and those assumptions are what made the plan work.

There are four ways in. You can enter a market that already exists and try to outperform what’s there, competing on features against incumbents who set the terms of comparison. You can resegment that market on price, taking the low end who’ll accept good-enough performance in exchange for paying a lot less. You can resegment it for a niche, serving a subset the incumbents ignore, sometimes at a higher price and often with worse performance on the attributes that subset doesn’t care about. Or you can create a market where your customers couldn’t do the thing at all before. Steve Blank laid this out in The Four Steps to the Epiphany in 2005. His categories are existing, resegmented, and new, with resegmentation splitting two ways. He reckons more than half of startups walk through one of the two resegmentation doors.

Each door changes the target, the argument, the clock, and the budget at once. Enter an existing market and you’re selling to the incumbents’ customers, who want better performance and will score you against a list they already have. Resegment on price and you’re selling to the cost-sensitive end of that same base, at the risk of the incumbent deciding to match you. Resegment for a niche and you’re telling a specific group that everything else on the market was built for somebody else, at the risk that the group turns out to be too small to live on, or not distinct enough to buy any differently from the mainstream. Create a market and you’re selling against people doing nothing at all, at the risk that adoption never arrives.

I’ve argued before that nobody finds a market; you build one or take a seat in somebody else’s. The four doors are what that choice looks like in operational terms, before anyone writes a plan.

The Door Is a Choice

Blank frames the four as a diagnosis, a question you answer about a market that already is what it is. Three things you control move a market from one type to another.

Price moves it. Come in far enough under the incumbent and what would have been an existing-market fight becomes a low-cost resegmentation, which brings a different buyer, a different motion, and a ceiling on what you can ever charge.

How wide you draw the problem does the same work. Narrow it to a subset nobody serves and you’re a niche entrant. Widen it past anything a buyer can currently purchase and you’re creating a market, whether or not you meant to.

The third lever has no slot in Blank’s scheme at all. It’s the buying center you enter through, meaning which person inside the customer you actually sell to. There are always at least two. One is whoever will use the thing every day. The other controls the budget and signs the contract. Blank’s four types quietly assume you’re selling to the same buyers the category has always been sold to.

The Fifth Door

Sell to the person who uses the thing rather than the person who signs, and the market type moves underneath you. Payments in 2011 was a thoroughly settled category, with incumbent processors, a known budget line, and a procurement process every finance team already knew how to run. Stripe sold to the developer, who had never been the buyer of anything. Everything on the seller’s side of that deal was ordinary existing-market business, and the buyer was somebody the category had never once pitched.

Those two halves want opposite plans, and running both at once is the part nobody budgets for. The category hands you named competitors, a price the market already understands, and a problem no buyer disputes is worth solving. That’s ordinary existing-market work, and at least it’s legible. You’re arguing you’re better than a specific alternative in front of people who agreed long ago that the category deserves a budget line. What you build for it is competitive positioning, a differentiation claim that survives a bake-off, and a clean migration path off the incumbent.

The buyer hands you somebody who can’t sign a contract, has never sat through a vendor pitch, and will make up his mind by signing up and trying the thing well before he talks to anyone. That’s new-market work, and it costs differently. He can’t be sold to in the ordinary sense, so the documentation becomes the pitch and the free tier becomes the demo, and the product has to close itself with nobody in the room. He also can’t buy. Adoption shows up before revenue does, and at some point you have to go find the person with the budget and sell her a tool her own staff installed months ago.

So you’re funding two motions at the same time, and in most companies they report to different people. The education work sits with docs and developer relations, measured on adoption. The competitive work sits with product marketing and sales, measured in pipeline. Each side can look at the other’s budget and see something optional, which is the argument that surfaces the first time you have to cut. Cut either one and the machine stops. Lose the education and nobody arrives; lose the competitive work and the people who did arrive can’t get a purchase order approved.

Two Quarters of Expensive Confusion

Most developer-tools companies live in this position. Datadog entered a monitoring market with entrenched incumbents, Twilio sold into telephony, a category older than software itself, and MongoDB went after Oracle. The category was settled in every case and the buyer was brand new, and the plans that come out of that combination look broken to anyone trained on enterprise go-to-market. The documentation is public and ungated, so no leads come off it. There’s a free tier that on paper eats revenue. Sales gets hired late and hired to expand accounts the product already won rather than to source new ones. The pipeline report reads thin, because the signal that matters is sitting in usage data the CRM never sees.

Drop a conventional enterprise plan on top of that and every change in it is defensible on its own. Gate the documentation to capture leads. Retire the free tier to protect the average selling price. Hire SDRs to call the person who controls the budget and buy the analyst placement while you’re at it. What breaks is the top of the machine. The practitioner who would have found you through the docs and tried you for free never arrives, and he or she was the entire acquisition motion.

It takes a couple of quarters to see because revenue lags badly here. Practitioners who already installed you keep converting into paid accounts on their own schedule, so the numbers hold while acquisition quietly stops. By the time the revenue line bends, the cohort that would have fixed it stopped showing up two quarters earlier, and the plan that caused it has been running long enough to look like the status quo.

One Team, Two Doors

Donna Dubinsky ran Palm Computing when the Pilot shipped in 1996 and was CEO of Handspring when the Visor shipped in 1999. The two go-to-market plans were opposites, and both were right. At Palm the job was convincing people a handheld was worth wanting at all because nobody was shopping for one. By 1999 buyers compared handhelds on specifications, and Handspring differentiated on expandability and performance in a market Blank describes as already worth a billion dollars. Handspring running Palm’s evangelism would have spent a year teaching buyers what they already knew, and Palm running Handspring’s feature comparisons would have burned its cash arguing against products nobody was cross-shopping.

Blank calls 1996 a market that didn’t exist, which is tidier than the record. Handhelds had been on sale since the Psion Organiser in 1984 and had simply never sold in volume. What changed between 1996 and 1999 was buyer literacy: whether people knew what the thing was for, what to compare it against, and which budget it came out of. Neither Palm nor anyone else bought that literacy; it accumulated over four years of people watching a colleague pull one out in a meeting.

Your Dashboard Can’t Tell You Which Door You Walked Through

The uphill direction of travel is running a new market on existing-market assumptions. You fund demand capture when there’s no demand yet to capture. You buy search terms nobody types. You staff a sales team against a buying process that hasn’t formed. None of it looks like waste while it’s happening, because every line item is something a competent company does.

The obvious correction is to run the plan and watch the numbers, adjusting when they disappoint. But early success in a new market looks just like failure in an old one. Low volume. Long cycles. Prospects who engage warmly and never convert. Read as an execution problem, those numbers call for more spend and more reps. Read against the right market type, the same numbers are on schedule, and the correct response is to protect the runway. No metric tells you which situation you’re in, and the dashboard will quietly recommend the wrong one.

Each door has its own objective, and the numbers only mean anything once you know which one you’re walking through. In an existing market the objective is share: a slice of spending that already exists, taken from somebody who currently has it. The work is capture against a buying process that already runs. In a resegmented market it’s share and education at once, which is harder than either alone and also the most common way in. In a new market there’s nothing to take and nobody to take it from, so the objective has nothing to do with share. The measure is whether the number of people who could plausibly buy is moving toward something a business could live on.

The Flattering Answer Is the Expensive One

Picking a door is supposed to be a judgment about the market. It’s also a judgment about which conversation you’d rather have with your board. “New market” buys a longer runway and more patience, and it’s a claim a founder can always argue for, because nobody can disprove it inside a year. “Existing market” obliges you to explain how you’ll beat a funded incumbent with a tenth of their go-to-market budget. The answer that’s easiest to defend is also the most expensive one to be right about.

Money settles what the analysis can’t. Add up what each door costs, set it against what’s in the bank, and where two types both fit, take the one you can pay for. Each carries a different bill. A low-cost resegmentation needs a cost structure that survives the incumbent matching your price, and they can usually hold a lower price longer than you can. A niche resegmentation needs a subset both large enough to live on and cheap enough to reach because a small audience can’t absorb a big acquisition cost. A new market needs enough runway to pay for the education before anybody buys. When the only type you can afford isn’t the one you want, you have a financing problem wearing a market-type label, and it belongs in front of your board as a financing problem.

The Door Closes Behind You

A market can be new when you enter it and ordinary three years later. Nothing announces the change when it comes. So you keep spending to teach people what the category is for long after they’ve learned it, and you’re still funding evangelism while your buyers have started comparing feature lists.

So market type belongs on the quarterly agenda, next to pricing and segment, which you already revisit. Ask whether you’re still in the market you planned for. If you aren’t, ask what your objective and your spending should have changed to and how long ago.

The longer you wait, the less of a choice it is. Price, how wide you draw the problem, and which buyer you sell to are all yours to set before launch and expensive to change after. A published price fixes what the market thinks you’re worth. A buying center you’ve sold into for two years has headcount attached to it, comp plans written around it, and a content library aimed at it. You can pick the type while it’s still on paper. Once you’ve spent two years executing it, you aren’t choosing anymore, you’re living with it.

Nobody “finds” a great market. You build one, or you take a seat in somebody else’s.

It’s easy to end up in someone else’s market. For example, “searching” for product-market fit casts you as an explorer hacking toward a golden idol already buried in the jungle: fit exists somewhere out there, and enough machete work will reveal it. Tweak the messaging, adjust the pricing, and a hungry crowd surfaces.

I’ve defended that search, and I still do. Steve Blank built a methodology on it. The Four Steps to the Epiphany (2005) describes a startup as a temporary organization looking for a repeatable business model, and running one as a small version of a big company kills it. The customer development loop the book introduced is now the standard for how founders talk to buyers before they build, and it deserves to be. I’ve made the case for accepting the search and for grounding that phase with execution metrics.

The Market You Choose Shapes the Frame

The Four Steps to the Epiphany says startups are either in existing markets, resegmented markets, or new ones.

Blank encourages founders to figure out which market type they’re in, as though it were something the market had done to them. It’s a decision made early, often accidentally, and then defended for years by everyone who joins afterward.

Confined to a market that already exists, most founders reach for segmentation to carve out a slice of it. Sort a population by firmographics and geography and you get a TAM slide the board will nod at. That slide describes the world as it is. Whether anyone in that world will listen to a new claim is a question segmentation has no way to ask.

When you pursue an existing segment, you inherit the buyer’s mental models along with it. You enter an existing budget line. You accept a price ceiling somebody else set. A database buyer measures you against Oracle. A CRM buyer measures you against Salesforce. You’re selling on ground somebody else surveyed, and your win-loss reports come back saying “feature gap” and “price” more often than you’d like.

Why the Funnel Can’t Fix It

The reflex at that point is to work the funnel. Rewrite the headline, rebuild the demo, run the test again. A buyer who hasn’t been given a way to read a new thing will supply one himself, and he’ll reach for whatever costs him the least thought: you’re a cheaper Oracle, or you’re a line item he isn’t going to defend to anyone. Every test you run gets scored inside that sentence.

Funnel work can’t reach the problem because of where it sits in the buyer’s decision. A funnel measures conversion inside a frame somebody already set, and the setting happened upstream, out of view. By the time anyone reaches your pricing page, the comparison set has been chosen, the budget line has been picked, and your champion already has a sentence she uses to describe you to her CFO. A better headline might move her through that decision faster while the decision itself stays closed.

So the gains are real and capped. Each round of testing tunes the message toward the buyers who already understood you in the incumbent’s terms, since those are the people converting and producing the data. Anyone who might have responded to a different story never appears in the results because nobody told them one. What you get is a well-tuned machine for recruiting the customers most likely to grind you down on price.

Building the frame instead starts with a specific claim about how the world has changed and uses that claim to redraw the context your product gets judged in. Positioning leaves the product alone; it changes the surroundings until your product is the obvious conclusion. Three moves do that work.

Three Moves to Build the Frame

A market doesn’t change because you call your product something new. It changes when buyers are given a new way to understand the problem, the moment, and the outcome. Building that frame takes three moves.

Name the Enemy

The enemy is the status quo: the friction your customer already resents and has stopped mentioning, because mentioning it never helped. It’s a way of working that has to end. Name a competitor instead and you’ve validated the competitor, then spent the rest of the sales cycle arguing inside its frame.

At Heroku, the market that already existed was obvious and terrible. Developers wanted somewhere to host Ruby apps, which meant a commodity knife fight against every VPS provider, settled on specs per dollar. We went somewhere else. Our enemy was infrastructure management, the whole body of sysadmin chores that had nothing to do with shipping software. Nobody was calling it an enemy at the time. Developers assumed servers were part of the job, the way you assume weather.

Explain Why Now

Name the change in the world that makes the old way untenable. This is the hinge, and it’s where designed markets mostly fail, because the founder reaches for a shift that turns out to be a wish.

A real one passes a test you can run before writing a word of copy: somebody who doesn’t work for you can verify it, and it happened before your pitch. If your why-now is “buyers are waking up to the need for this,” you don’t have one; you have a hope with a date attached. If it’s a change in what something costs, or a capability that shipped and didn’t exist eighteen months ago, a skeptic can check it.

Heroku’s was sitting in public. Rails had made web applications fast to write, and EC2 had made servers rentable by the hour since 2006. The distance between how quickly you could write an application and how long it took to get one running had grown absurd, and anyone paying attention could see it. We never had to convince a buyer the shift was real. We had to name what it meant.

Describe the Promised Land

Sell the new way of working. Features are evidence for it, and they only get interesting once the buyer wants the world you’re describing. Latent irritation becomes a market requirement when somebody names it out loud and gives buyers the language for it.

Heroku’s promised land was that you deploy code and stop thinking about servers. We made time-to-deploy the measure, and the incumbents had no answer for a yardstick they’d never thought to compete on.

A yardstick only counts once it leaves your marketing. Ours left through documentation. In 2011 Adam Wiggins, one of Heroku’s founders, published the Twelve-Factor App, a plain description of how an application should be built so it can be deployed and scaled without ceremony. It read like engineering guidance instead of positioning, which is precisely why it traveled. Teams that never bought anything from us adopted it. Competitors built products against it. The assumptions underneath it, disposable processes and config kept in the environment and no server anyone nurses by hand, became the default way a generation of developers thought about running software. By the time the category had a name, Platform as a Service, the argument had been settled somewhere upstream of any sales call.

Salesforce ran the same play on a bigger stage. The market that existed was people who wanted cheaper on-premise CRM. Marc Benioff’s company launched its on-demand service in February 2000 and sold it against packaged software itself: the eighteen-month implementations and the upgrade cycles and the IT tax. The web had made hosted software workable, and that was the why-now. The promised land was that you log in and start working. The red “no software” circle was a thesis compressed into a logo.

I’ve written about Mesosphere at length elsewhere. The part worth adding here is the pricing. Selling a Data Center Operating System instead of a cluster scheduler moved us against a different budget line, with a different approver and a much higher ceiling. The reframe held for about eighteen months, and then Kubernetes was free and we weren’t. Winning the argument about what a thing is doesn’t settle who gets paid for it.

The Play Isn’t Confined to Software

Dietrich Mateschitz found Krating Daeng, a Thai tonic sold to truck drivers and laborers, and saw a product with no Western market. By Red Bull’s own account of its founding, the market research he commissioned came back brutal: testers disliked the taste, the logo, and the name. Those are reasonable objections if the thing on the table is a soft drink.

Mateschitz changed what was on the table. He reformulated and carbonated it, launched Red Bull in Austria in 1987, priced the slim can well above cola, and seeded it with students and extreme-sport athletes. The price did argumentative work, since nothing costing triple a Coke is ordinary refreshment. The users did the rest, demonstrating in public what the thing was for. “Energy drink” became a Western category because Red Bull handed buyers a way to understand a product that had been sitting in Thailand for years.

What the Bet Costs

A designed market is a falsifiable bet on reality, and that’s what makes founders flinch the most. Slot a product into an existing category and you can post mediocre numbers for years without ever learning whether the premise was wrong. Every miss gets charged to execution, so you hire two more reps and run the quarter again. Design a market and the verdict arrives early and in public. You’ve named a shift in the world, built the pitch on it, priced against it, and told everyone what to expect. If the shift isn’t real, buyers won’t argue with you; they’ll just fail to show up, and that silence is legible to your board and to every competitor watching.

There’s a line here worth being honest about, because from the outside the move looks identical to spin. Reframing works when the product can actually deliver on the new measure. Heroku could genuinely take deployment down to seconds. Red Bull genuinely did something to you that a cola didn’t. Move the yardstick onto ground you can’t hold and buyers find out in their first quarter of use, and you’ve taught the market a vocabulary that now indicts you. The frame has to be true before it’s useful.

Keeping the bet honest takes the same discipline the search phase demands: write down what has to be true, decide in advance what evidence would change your mind, and hold yourself to the list. What changes when the market is designed instead of inherited is the exposure. Your hypothesis stops being private. It’s on your home page, and your competitors have read it.

The other cost is that you can’t hand this off. The organizational temptation is to push go-to-market downstream to a VP of Sales and get back to the product. But a VP of Sales optimizes inside whatever frame he inherits, because that’s the job you hired him to do. Deciding what the company means sits outside that job, and it stays with the founders. Two questions get you started, and you have to answer them about your own technology: who loses if we win, and what happens to our customer if they don’t buy from us. No feature list has ever answered either one.

Slotting into an existing category usually does get you revenue faster, and it locks you into the incumbent’s price. That trade looks like traction on a monthly board update and reads as a race to the bottom two years later.

None of this makes the outcome safe. Mesosphere named its category, held the frame for a year and a half, and lost the market anyway. What the work buys is narrower than the headline examples suggest: a market that’s yours to lose rather than somebody else’s to defend. Whether that trade is worth making is the founder’s call, and it has to be made while the pitch is still being written.

Subscribe now

Most CMOs aren’t losing their jobs to AI. They’re losing them to the data layer.

That’s the less dramatic version of the story, and the one more likely to be true. The tools aren’t the threat. What the tools reveal is the threat. Point a capable AI system at a marketing org and it will find, within about a week, every place the machine underneath is held together with duct tape and quarterly heroics. Then it builds on that foundation, faithfully and fast, and hands you back your own dysfunction at ten times the volume.

The work we judge marketing on hasn’t changed much: the message, the brand, whether the pipeline number lands at the end of the quarter. That still gets you hired. It won’t keep you in the seat. For most of the last two decades, a CMO could be brilliant at the visible half of the job (positioning, story, the launch) and treat the plumbing as somebody else’s department. Data lived with IT or a RevOps team. Attribution was a slide you argued about once a quarter and then stopped thinking about. Experimentation, if it happened at all, was whatever the agency felt like testing that month.

That arrangement is quietly ending, and AI is the reason.

When execution gets cheap, the bottleneck moves

AI makes marketing execution cheap and fast. Generating a hundred ad variants, personalizing a lifecycle flow, drafting landing pages, scoring leads, forecasting spend: all of it collapses in cost. When execution gets cheap, the bottleneck moves to the quality of the inputs and the speed of the learning loop. If your customer data sits in twelve systems that don’t talk to each other, no model fixes that for you. It averages across the gaps and produces confident, well-written, wrong output. Bad data doesn’t make a bad report you can ignore. It trains bad targeting, teaches the system the wrong lesson about who to chase, and ships forecasts built on holes, all of it automatically.

The companies already run this way

Look at the companies that already run marketing this way. Booking.com has said publicly that it runs more than a thousand concurrent experiments at any given moment. Amazon built an internal experimentation platform, Weblab, because it decided the rate of learning was the thing to optimize, not any single campaign. Netflix treats its recommendation and merchandising systems as a testing surface rather than a set-and-forget feature. None of these companies won by hiring a more poetic copywriter. They won by building a system that turns a market signal into a decision faster than their competitors can, and by keeping the data clean enough for that system to be trusted.

Now put an ordinary B2B marketing org next to that. The CRM is half-empty because sales never had a reason to fill it in. Product usage data lives in a warehouse marketing can’t query without filing a ticket. The attribution model gets re-litigated in every QBR because nobody actually believes it, so nobody wants to bet a budget on it. Three teams have three definitions of a “qualified” lead. This isn’t a tooling problem. You can buy the best martech stack on the market, drop it on top of that mess, and the mess wins. It always wins.

The job description is being rewritten

So the job description is being rewritten, whether or not the titles catch up. Knowing the customer, owning the story, running campaigns, minding the brand: table stakes now, not the whole hand. The CMO increasingly has to answer a different set of questions. What data does the company actually capture, and which of it does anyone trust? What’s connected to what? How does a signal from the market become a decision, and how many weeks does that take? Where do the feedback loops between creative, sales, product, service, and finance break? Those are architecture questions, and most of us were never trained to answer them, because for most of our careers we didn’t have to.

A lot of marketing leaders came up learning to sit above the system rather than build it. We were rewarded for taste, for narrative, for the big launch, and we treated the data layer as back-office hygiene, the kind of thing you delegate and forget. In an AI-run marketing stack, the back office is the strategy. The quality of your data governance determines the quality of everything the models produce downstream. Your ability to run clean experiments determines how fast you can learn. Measurement stops being a reporting function and becomes the steering wheel. If you can’t defend your measurement model in a hostile finance review, you can’t defend your budget, and pretty soon you can’t defend your role.

The craft becomes the scarce part

None of this means the craft stops mattering. It matters more, because it’s now the scarce part. When anyone can generate a competent campaign in an afternoon, competent stops being a differentiator. Taste still separates the good from the forgettable, and so does the judgment to know which idea is worth betting a quarter on. Plenty of brands have won on story alone with a mess behind the scenes, and some still will; that window is just closing as the mess becomes computable. The craft now sits on top of a foundation a modern CMO has to own rather than delegate: data architecture, experiments you can design and defend instead of merely approve, a real point of view on what the models are allowed to touch and how you’ll know when they’re wrong.

Do the audit before AI does it to you

So do the audit AI would do to you, before it does it. Map where your customer data actually lives and how much of it you’d stake a forecast on. Count how many genuine experiments you ran last quarter, not how many “tests” someone name-checked in a deck. Ask whether your attribution model would survive an hour with a skeptical CFO. The answers tell you how ready your function is to move at the speed the tools now allow, which is a different question from how many tools you’ve bought.

The CMO who comes out of this ahead owns one number above the others: how fast the company learns from its market. It’s an unglamorous number. Nobody puts it on a launch slide, and no agency can move it for you; it lives in the plumbing most of us spent a career delegating. Claim it and the tools multiply everything you’re already good at. Leave it unclaimed and they’ll multiply the mess instead, at ten times the volume, in your name.

Subscribe now

Two marketing teams get the same tools on the same Monday. Same models, same budget for tokens, same access to the same vendors pitching the same “AI-native” platform. Six months later one team is shipping sharper work in half the time and the other is drowning in competent, forgettable output. The tools didn’t cause the gap. They revealed it.

This is the part the vendor demos won’t tell you. AI is not a skill you bolt onto a marketing org. It’s a mirror. Point it at a team with a muddy point of view and a fuzzy sense of who the buyer is, and it produces muddy, fuzzy work at ten times the rate. Point it at a team that actually knows its market, and it amplifies that clarity into reach the team could never have hit by hand. The technology is neutral. The marketer is not.

The lens that beat the specialists

The marketers who are hardest to compete with don’t win on a channel tactic. They win on a habit of mind, often carried in from somewhere outside marketing entirely, like manufacturing or supply-chain work: they look at a business as a system of interacting parts. A marketing operation, seen that way, is a production engine assembled from components. Pricing feeds positioning. Positioning feeds the message. The message feeds the pipeline. Sales feeds back what the message is actually doing in the wild. Change one part and the others move.

Most marketers are trained to do the opposite. The advice you’ve heard your whole career is to specialize: become the best paid-search operator on the planet, or the best lifecycle-email person, or the best brand storyteller. Deep specialization buys you mastery of one slice and blindness to the rest of the machine. The paid-search expert optimizes a campaign that’s selling the wrong promise to the wrong segment, and the dashboard looks great right up until the pipeline doesn’t convert. The specialist did the job. The system still lost.

Here’s why that matters more now than it did five years ago. The specialist’s slice is exactly what AI is best at. Bid management, keyword expansion, variant generation, first-draft copy, the assembly of an attribution report: this is information processing, and information processing is precisely what the machine automates first. If your value to a marketing org was that you were the person who could operate one platform better than anyone else, the platform now operates itself well enough. The premium on that skill is already falling, and it isn’t coming back.

Where the value goes instead

So where does a marketer’s value migrate? Not to whoever prompts the fastest. It migrates to judgment, the one capability AI structurally cannot supply.

A model is a prediction engine. It surveys everything that has already been written and predicts the most reasonable average next move. That’s enormously useful and also its hard ceiling. Ask it to position a genuinely new category and it will hand you the existing language for the nearest old one, because the new framing isn’t in the training data yet. Someone has to originate it. Ask it whether to plant your flag on “data observability” or invent a term the market doesn’t have words for, and it will give you a tidy list of pros and cons. It cannot tell you which bet is right for your company, your moment, your tolerance for being misunderstood for eighteen months. A human still has to decide. Category creation is the purest version of this: by definition you’re naming something the corpus has no consensus on, so the corpus can’t name it for you.

That’s the work that’s becoming scarce, and scarce work is where the money goes. The marketer who’s worth keeping isn’t the one with the most certifications. It’s the one with the clearest read on the buyer and the steadiest hand on the bet.

Judgment is a muscle, and it’s atrophying

There’s a quieter problem underneath this, and it’s worth naming because it’s happening inside good teams right now. Judgment behaves like a muscle. Used, it strengthens. Unused, it wastes away. And the easiest thing to do with an AI is to hand it the decision.

Watch how a content brief actually gets made now. A marketer feeds the model the source material and asks for the angle, the headline, the structure. The output feels like progress, so the habit compounds. What got quietly outsourced wasn’t the typing. It was the interpretation, the part where a person reads the market and forms an opinion about what to say. Do that for a quarter and you still have a content team. Do it for two years and you have a content team that can no longer tell a sharp angle from a safe one, because the muscle that made that distinction hasn’t been loaded in a long time.

The fix isn’t to ban the tool. It’s to change the job you give it. Use the model to attack your thinking, not to replace it. Form your own read of the positioning first, then hand the machine the argument and tell it to find every hole: the counterexample, the segment you’re ignoring, the reason a skeptical buyer would bounce. There’s documented work, including a Procter & Gamble study run with Harvard Business Review, showing that a person paired with an AI used as a sparring partner performs at roughly the level of a whole team. The order of operations is the entire game. Decide, then pressure-test. Reverse it and you’ve trained yourself out of deciding.

From a map to a compass

Most marketing leaders were trained to run on a map. Build the annual plan, commission the research, chart the route, then follow it with small corrections quarter to quarter. That worked when markets moved slowly enough to forgive the lag. It’s now a liability. A category window can open and close inside the six months it used to take to validate the move, and a plan that assumes a stable market is a plan that’s wrong on arrival.

What the work demands instead is a compass: the ability to move through terrain you don’t have a map for, read how the ground is shifting, and change direction without waiting for permission from a deck. A compass needs a fixed point to orient against. For a marketing org, that fixed point is positioning, the bedrock answer to who you’re for and why you’re the obvious choice. Get that right and you can improvise everything downstream, because every tactical call has something true to test itself against. Get it wrong and no amount of AI throughput saves you. You’ll just be lost faster, with better graphics.

The hype keeps insisting the winners will be the teams that adopt the most. The opposite is closer to true. The winners will be the ones who know what they’re doing well enough that the machine has something real to amplify. AI made execution cheap. It made taste, judgment, and a defensible point of view the only things left worth paying for. If you’ve spent years building those, this is the best market you’ll ever work in. If you outsourced them to the tool, you already have your answer to why the other team is pulling ahead.

Subscribe now

For most of my professional life, a website’s job was simple. A person landed on your homepage, read the headline, clicked through the nav, compared a couple of pages, and if you’d done your work, filled out a form. Every instinct we built as marketers points at that person: the headline that earns the scroll, the proof that earns the click, the call to action that earns the email address.

That person is still showing up. But more and more, they send someone ahead. An AI agent reaches your site first now, working on the buyer’s behalf. It summarizes vendors, builds a shortlist, drafts the internal recommendation, and decides whether you belong in the conversation before a human opens a tab. Sometimes the human never visits at all. They read what their agent told them about you, and they act on that.

This moves the job. Your website used to be a persuasion surface pointed at one reader. Now it’s also a reasoning surface that a machine mines for structure, and those two readers want different things from the same pages.

Here’s the line I’d put on the whiteboard: your website is becoming your company’s public prompt surface. It’s the raw material an agent uses to assemble an answer about you, for someone, while you’re nowhere near the room. Most sites make a poor prompt. They’re abstract, they’re inflated, and they bury the few facts an agent actually needs under copy engineered to make a human feel something. Closing that gap is the work.

For an operator who has spent a career optimizing for the human funnel, this is uncomfortable, because the new reader is unmoved by everything that used to convert. So let’s redesign for it.

Every Page Is the Front Door

Humans may still enter through the homepage. Agents rarely will.

An agent meets your company in fragments: a search result, a docs page, a pricing table, a GitHub readme, a LinkedIn post, a podcast transcript, a scraped snippet, a PDF someone uploaded into a chat window three years after you published it. It assembles its picture of you from whatever surfaces first, in whatever order, with no guarantee it ever sees the page you consider the real introduction.

So every page that matters has to stand on its own. Each one should answer the same short interrogation without relying on the page before it: What is this company? Who is it for? What problem does it solve? When should someone recommend it, and on what evidence? What’s the next step? A page that only makes sense after you’ve read the homepage is a page an agent will quote out of context, or skip.

That pushes the site away from a linear brochure and toward something closer to a knowledge graph, where each node is a self-contained explanation of the company, the category, and the buyer’s path through it. The working rule is blunt: assume every page is the first page an agent sees, because for someone, it will be.

Write Claims a Machine Can Lift

Most marketing copy is useless to an agent because it’s abstract, inflated, and interchangeable. “We help teams unlock the future of intelligent transformation” sounds finished and says nothing. An agent can’t classify it, can’t compare it, can’t quote it into a recommendation. It’s noise that happens to be grammatical.

Now picture the alternative. “We help enterprise support teams cut ticket resolution time by triaging requests, retrieving policy-compliant answers, and escalating the hard cases to human specialists.” That version hands the agent a category, an audience, a problem, a mechanism, and a use case. It survives extraction. Lift any sentence out and it still means something.

This is where category work and machine-readability turn out to be the same discipline. The vocabulary you use to describe yourself is the vocabulary an agent will repeat when it describes you to a buyer. If your language is vague, the agent paraphrases you into the generic middle of your category, indistinguishable from the four other vendors it’s also summarizing. If your language is precise and owned, the agent carries your terms forward intact. Controlling the words a machine uses for you is the linguistic moat, rebuilt for a reader that takes you literally.

Declarative statements travel best. “We’re best for.” “We’re not a fit for.” “Teams use us when.” “The main alternatives are.” “The strongest proof is.” None of this dumbs the site down. It makes the company legible, which is a harder thing to be than clever.

Build the Layer the Buyer Can’t See

Underneath the site humans look at, you want a layer built for the machine. It doesn’t have to hide from people, but it isn’t designed to charm them. It’s designed to be retrieved.

The ingredients are familiar to anyone who’s done technical SEO, just pointed at a new consumer: an llms.txt file, schema markup, structured metadata, clean docs, a canonical set of company facts, a product taxonomy, comparison and objection pages, a plain pricing explanation, security and compliance documentation, machine-readable FAQs, citation-ready proof. Most of it you already half-own. The shift is treating it as a coherent layer rather than scattered exhaust.

The single highest-leverage piece is a plain-language briefing written directly at the agent. Call it an Agent Briefing or a Vendor Brief for AI Assistants; the name matters less than the candor. It should be structured, direct, and boring in the best possible way:

Company:
Category:
Primary customer:
Primary buyer:
Primary problem solved:
Best-fit use cases:
Not a fit for:
Main alternatives:
Key differentiators:
Proof points:
Security posture:
Pricing model:
Evaluation path:
Recommended next step:

The discipline of filling that in does something useful even before any agent reads it. It forces the company to say plainly what it actually is, which is a sentence a surprising number of companies can’t finish.

And while you’re at it, reorganize how you think about pages. A conventional site is built around sections: Home, Product, Solutions, Resources, About. An agent-facing site is built around the questions a buyer types into a chat window. What does this do? Who’s it for? When should I not use it? How’s it different from the obvious alternative? What does it cost? How hard is it to adopt? What would a technical evaluator pick at? What proof exists? What’s the fastest way to test it? Each of those deserves a clear, structured answer living somewhere findable. The goal isn’t a content swamp. It’s a site that can be answered.

Tell the Agent Who You’re Not For

Two kinds of pages most marketers treat as afterthoughts become load-bearing in an agent-mediated market: the comparison page and the disqualifier.

Comparison pages are usually thinly disguised hit pieces, and an agent reads right through them. A page that says “us good, them bad” gives a machine nothing to reason with, so it discounts the whole thing. A page that explains the actual tradeoff gives it material it can use. What is each approach genuinely good at? Where do they overlap? When should a buyer pick one, and when should they run both? Write it as a decision, not a verdict. “Choose this if you need X. Choose that if you need Y. Run both if your architecture forces Z.” That page earns trust from the human and hands the agent honest inputs, which is the only kind it can safely repeat.

The disqualifier is harder, because it runs against every instinct we have. Companies are afraid to name who they’re not for. In a market where an agent is deciding fit, that fear costs you. A good agent isn’t trying to find the most enthusiastic vendor; it’s trying to find the right one, and it needs boundaries to do that. So give it some. “This probably isn’t a fit if you only need a prototype, have no production reliability requirements, or could solve the problem with a short internal script.” Naming the wrong-fit buyer doesn’t weaken your positioning. It sharpens it, and it keeps the agent from recommending you into deals you’d lose anyway.

Make Your Proof Portable

Humans skim proof. Agents retrieve it and reuse it, which means your proof points have to survive being lifted out of your site and dropped into someone else’s memo.

“Trusted by leading teams” survives nothing. It carries no fact. “Used by enterprise security teams to cut manual review time on high-volume access requests” carries a context an agent can place. And a fuller version carries more: a deployment window, an organization size, a before-and-after number, a note on what stayed under human control. The richer the context attached to a claim, the more accurately a machine can repeat it without inventing the parts you left out.

Not every company can publish hard metrics, especially early, and you shouldn’t fabricate them to fill the template. But every company can make the proof it does have more legible: quote a named customer instead of an anonymous one, attach a number to an outcome, document the security posture, show the architecture. The test is portability. A buyer’s agent should be able to move your proof into a recommendation without guessing what it meant, because the guess is where you get described badly.

Write for the Memo You’ll Never See

Most B2B sites optimize for conversion. The page that matters most in an agent-mediated deal is one you’ll never design, because the buyer’s agent writes it: the internal recommendation that goes to the team.

That memo has a predictable shape. What the company does. Why it matters now. Where it fits in the stack. What problem does it solve that the current approach doesn’t. Who should evaluate it. What the risks and objections are. What evidence backs it. What the next step should be. If your site supplies those pieces cleanly, the agent assembles a credible case for you. If it doesn’t, the agent either leaves you out or describes you in a way you’d never sign off on, and you never find out it happened.

So write toward transmission, not just conversion. The question isn’t only “will this page persuade a visitor.” It’s “can a buyer’s agent turn this site into an internal recommendation that survives scrutiny from people who never visited it.” Optimizing for the memo you’ll never see is the closest thing this market has to a new fundamental.

Keep the Human Side Beautiful

None of this is an argument for an ugly, robotic, utilitarian site. The opposite. As the machine layer absorbs the facts, the human layer is freed to do the one thing the machine can’t: create belief.

The human visitor doesn’t need every FAQ, integration, and objection stacked on the homepage. They need orientation and confidence. They want evidence that the company has taste, judgment, and a coherent read on where the world is going. They want a thesis, the stakes, the mechanism, the proof, and a clear next step, delivered with enough conviction that they trust the people behind it. “Teams are adopting AI faster than their operating models can absorb, and the result isn’t just more automation; it’s more decisions, more handoffs, and more ways to get burned. We give teams a governed way to put agents into real workflows without losing visibility or accountability.” That gives a person a point of view. The agent layer underneath supplies the receipts.

Think of the visible site as a well-lit gallery and the agent layer as the archive and catalog behind it. The companies that win this won’t choose between the two. The human side earns belief; the machine side earns distribution, the quiet kind that happens inside chat windows and recommendation memos you never see. You need both, and they reinforce each other when you stop pretending one reader can do the job of two.

The Clever Move: Copy Briefing for My AI

Here’s the one new mechanic I’d ship first, because it’s cheap and it signals that you understand where buying is going.

Put a button on your key pages that reads “Copy briefing for my AI.” When the buyer clicks it, you hand them a clean, structured summary of the company, product, fit, alternatives, and next step, ready to paste into whatever tool they already trust:

I am evaluating [Company Name].

Summary:
[Company Name] is a [category] for [audience]. It helps [buyer/user]
solve [problem] by [mechanism].

Best fit:
Use it when [situation].

Not a fit:
It's probably wrong when [situation].

Key differentiators:
1. [Differentiator]
2. [Differentiator]
3. [Differentiator]

Main alternatives:
[Alternative categories or products]

Questions to ask:
1. [Evaluation question]
2. [Evaluation question]
3. [Evaluation question]

Recommended next step:
[Demo, trial, technical evaluation, docs, pricing request]

The old site asked the human to fill out a form so you could start selling. This one helps the human brief their agent so the selling continues after they leave, in language you wrote instead of language the agent guessed. That’s a small button doing real work.

A Short Audit

Before you publish, run the site against the reader who can’t see it. A handful of questions sort the legible companies from the rest:

• Can an agent name your category in one sentence and your buyer and user in two more?

• Does every page that matters stand alone, or do half of them only make sense after the homepage?

• Are your important claims specific enough to lift and quote without distortion?

• Do you have the machine layer: structured metadata, schema, an llms.txt, a plain-language agent briefing?

• Could a buyer’s agent build a fair vendor summary, compare you to the alternatives, surface the real objections, and recommend the right next step, all from your site?

• And is the human experience still clear, confident, and worth a person’s attention?

If the answer to most of those is no, you don’t have a content problem. You have a legibility problem, and the machine reading your site right now is already acting on it.

The New Bar

The old website was a destination. The thing we’re building is a source: for humans looking for confidence, for agents looking for structure, and for buyers trying to justify a decision to people who will never visit the page.

The job is no longer just to persuade one person to click. It’s to make the company easy to understand, easy to compare, easy to trust, easy to cite, and easy to recommend by a reader that takes you at your word and repeats exactly what you gave it.

Design for the human. Structure for the agent. Win the shortlist before the buyer ever fills out the form.

Subscribe now

The old playbook had a comforting shape. You found an underserved segment, spent eighteen months building something proprietary, locked customers into multi-year contracts, and let switching costs do the rest. Defensibility was a thing you could point to on a slide: years of engineering, a data advantage nobody could clone, an installed base too heavy to move. Marketing’s job was to narrate that fortress and price the rent.

That fortress is rubble. A product that ships Monday gets cloned by Wednesday because the hard part of building it has collapsed into a weekend of prompting against a frontier model. The technical lead that used to last years now buys you a few months, sometimes weeks. Cursor reportedly crossed $100 million in annual recurring revenue in roughly a year, and a dozen credible competitors showed up while it was still working out its own onboarding. Once the product stops being the moat, every assumption GTM was built on needs a second look, starting with the most flattering one, that a great product markets itself.

So the real question for whoever owns marketing now is what’s actually left to defend. The honest answer is uncomfortable, because it sounds like the soft stuff we spent a decade learning to distrust. Two things are left standing: taste and velocity. Not the product, not the data, not the contract. The feel of the thing and how fast you can make the next one. That changes what positioning, category work, and demand gen are even for.

Taste is the cover charge

When the technical differentiators dissolve, everyone reaches for the same word. Taste. If the models are commoditized, the argument goes, then judgment becomes the moat: the aesthetic sense, the editorial restraint, the instinct for what’s right that a machine can’t fake. It’s a flattering story for people who do positioning for a living, which is precisely why I distrust it.

Paul Graham made the observation that undercuts it in an essay about why he likes Swiss watches. Good design is centripetal. It pulls everyone toward the same right answer. When most people have bad judgment, the one operator with taste looks like a genius. But taste isn’t a secret. It’s a skill, and skills diffuse. The tools that bake in good defaults are in everyone’s hands now, the patterns are written down, and the median product in any category gets better-looking every quarter. The field converges.

Which does something specific to your positioning. Taste stops being a differentiator and becomes the price of admission. A product that feels clunky is dead before the first demo. A product that feels great no longer wins on feel because the other three vendors in the bake-off feel great too. Linear is the example everyone cites, and it cuts both ways: its design didn’t just lift Linear, it reset what every buyer in project management now expects on day one, which means Linear’s own polish bought it less room than the case studies imply. When everyone at the table has taste, taste gets you to the table. That’s all it does. The deck that opens with “beautifully designed” is describing the floor and calling it the ceiling.

Velocity is the engine

If taste only gets you in the room, velocity is what keeps you there. Not launch-week speed. The kind that compounds.

The fashion creative director is the comparison most people reach for. They invoke it to mean “have a vision.” The actual lesson is that it’s one of the most precarious jobs in business. Houses burn through creative directors with a brutality that has nothing to do with talent because talent was never the question. The question is whether you can deliver the next collection on time, before the room’s attention drifts. Nobody’s grading the show that just closed. They’re grading the market’s belief that the next one will be good.

Software runs on that same clock now, only compressed. The investor Elad Gil has put it plainly: in AI, a feature lead that once meant years means a quarter. Build one excellent thing, try to harvest it for a decade, and you’ll watch the harvest get cloned before the renewal lands. Velocity is the operational and cultural muscle to keep shipping reinventions while the last one is still getting applause. Feeling the market move, deciding what’s next, getting it out the door before the feeling goes stale.

For GTM that’s structural, not a matter of tempo. A launch motion designed to fire twice a year can’t carry a product that changes in meaningful ways every six weeks. The messaging has to be re-derived as fast as the product moves, which means the positioning is never finished, and the category story underneath it has to be sturdy enough to survive constant churn in the feature set above it. The teams winning here run marketing like a collection, not a campaign. New drop, new beat, same point of view. For twenty years the hard part was building. Now it’s not stopping.

Brand is real. The ground under it isn’t.

The Swiss watch story is the one the luxury crowd loves, and it’s worth telling straight, because the ending always gets dropped. It ran in three acts over about forty years. First, engineering merit, when a Swiss movement was simply the most accurate way to know the time. Then the quartz crisis, when cheap Japanese quartz made accuracy a commodity and gutted the function the whole industry sold. Then the brand age, when the survivors stopped selling timekeeping and started selling status, heritage, the object on your wrist.

Software is running the same three acts at ten times the speed. A product launches with a genuine technical edge. Improving models commoditize that edge inside a quarter. And it’s shoved, fast, into competing on experience and feel and brand, the same pivot the watchmakers took a full generation to make. The optimistic reading of that arc stops at act three: build a brand, reach the harbor, charge the premium forever.

The harbor isn’t safe, though. A brand is only as durable as the ground it stands on, and the Swiss watch brand age held right up until the Apple Watch colonized the wrist. The threat was never a rival watchmaker with cleverer branding. It was a different device that reset the whole stack and erased the distinction the brand existed to signal. In AI, the ground never stops moving. The model under you, the interface, the device, and the basic shape of the interaction, any of them can reset inside the life of a single company. A brand built on one interaction pattern goes stale the day that pattern changes. The luxury comparison leaves this part out because luxury, for a long stretch, got to stand still. You don’t get that.

What’s actually left

Taste is the cover charge. Velocity is the engine. Brand is shelter on ground that won’t hold still. What compounds underneath all three is harder to name and harder to copy: the instinct for finding the next real problem before anyone’s paying you to.

The operators who last won’t be the ones with the most refined taste or the loudest demand-gen machine. They’ll be the ones who get restless with their own wins, who can smell the commoditization coming for their own product before the market prices it in, who are already chasing the next genuine customer need while the current one is still on the board. That sits upstream of every positioning exercise you’ll ever run. You can’t position your way out of a problem that’s stopped mattering to anyone.

There’s a real irony in here for those of us who build brands for a living. The way to build one that lasts is to stay willing to walk away from it. You protect a brand age by refusing to get comfortable in it. So the move isn’t to cosplay the creative director’s taste. It’s to build a GTM org that can survive the creative director’s actual job: own as much of your own infrastructure and distribution as you can, treat every product and every campaign as this season’s collection rather than a monument, and keep your judgment pointed at hard customer problems instead of at curating how the last answer looks.

From the cheap seats, the company that wins the next decade will look like a luxury house. Impeccable taste, frightening pace, a point of view people pay just to stand near. Up close it’s less glamorous than that. It’s a room full of people who’ve made their peace with the fact that nothing they ship gets to be permanent and who’ve stopped wishing it would.

The most dangerous move a marketing or sales leader can make is to drift away from the work itself. It happens slowly and it looks like a promotion. You stop building campaigns and start reviewing them. You stop writing the email and start approving it. One day you’re the person who hasn’t logged into the ads account in eight months, the one who reads a dashboard and tells the team to go execute and you’ve quietly become a different kind of employee than the one you used to respect. Some leaders formalize the drift by renaming themselves. “Evangelist” is the tell: a title that means the job is now appearances and pronouncements with nobody underneath you who’d report that you still do anything.

The player-coach

The alternative has a name too, borrowed from sports. The player-coach finishes a meeting and then opens the Google Ads account to adjust live campaigns. She reads the metrics and acts on what they say instead of glancing at a number and delegating the thinking. She knows which landing page the traffic is hitting because she pulled it up this week not last quarter. Staying tactical like this isn’t nostalgia for the entry-level grind and it isn’t a refusal to delegate; it’s the only reliable way to keep learning and it’s how strategy actually improves. The plan changes because you’re doing the work, because the next move only becomes visible up close. You can’t see it from the altitude of a quarterly review.

That distance used to be a perk of seniority. Now it’s a career risk. The channels and tools turn over faster than any leader can absorb secondhand and the people getting replaced are the ones who lost touch with execution and never noticed. The title shift already underway in consumer brands points at where this goes: when Pepsi traded its chief marketing officer for a chief growth officer that wasn’t cosmetic, it was a company deciding the person at the top of marketing needs to own a number and know how it moves. The same logic is arriving in software businesses. The framing is almost Darwinian. If you can’t run at the pace of your fastest operators you fall behind, and the way to stay in shape is to keep doing the reps.

You can’t audit what you can’t do

The real exposure isn’t a stale skill set. It’s the moment you can’t explain why leads from a given source stopped converting and you’re the one accountable at the table. That gap shows up most painfully when a company is spending money it has no way to evaluate. Plenty of organizations hire an agency, a consultant, or an in-house specialist and have no idea whether the work is any good because nobody in leadership knows how to do it themselves. One company had a single specialist managing a million dollars in media spend with no one above able to judge whether it worked. Another had eight million dollars in Google AdWords run by an outside agency; the chief executive opened the first call by admitting he was spending eight million dollars and didn’t know whether a dollar of it produced anything. That isn’t a budgeting problem. It’s a knowledge problem wearing a budget’s clothes.

Whoever owns the spend under whatever title has to be able to audit the people spending it at a detailed level. That requires real fluency in the craft not a vocabulary for sounding fluent in a meeting. The same principle is why the marketing leader should be the face of the company’s podcast rather than handing it to a junior producer. The finished recording is the least valuable part of it. The value is in the process: meeting the guests, choosing the topics, working out how the thing gets distributed and who actually listens. That’s where the learning lives and where the relationships that compound into pipeline get built. Delegate the podcast and you’ve delegated exactly the part worth keeping.

The small, ordinary practice

None of this asks for heroics. The practical version is almost embarrassingly small. Take the hours you’ve freed up as you’ve moved up and reinvest a slice of them into hands-on familiarity. Open the ads account and poke around. Pull up the landing pages you’re sending paid traffic to and ask whether you’d convert on them. Sort a campaign’s ad groups from most expensive to least and follow the money into whatever you find. It really is that ordinary and it’s the difference between a leader who can interrogate a number and one who can only receive it.

When the work corrects the plan

Doing the work also changes what you build because the tactical view keeps correcting the strategic one. Consider the webinar, run almost everywhere as a lead-generation machine. Spend a few weeks actually running them and the premise falls apart: audiences don’t want to be sold, won’t wait twenty minutes for substance, and tune out the company-slide preamble. Sit in the data and you stop treating the webinar as a list grab and start treating it as a brand asset that earns trust, which is the one thing a fast-growing startup with money but no reputation can’t buy. You’d never reach that conclusion from a strategy deck. You reach it by watching real people drop off at minute nineteen and asking why.

The same correction runs through content syndication where companies hand a vendor an ebook to re-skin and embed across other people’s blog posts, usually against broad top-of-funnel search traffic the company could capture on its own. A leader who has ranked a page for a keyword knows this in their hands and knows the audited win rate on those syndicated leads sits near one-tenth of a percent. That’s paying a middleman to waste your sales team’s time. The person who can see it clearly is the one who has done the work the vendor is charging to replace.

Show, don’t recommend

There’s a quieter benefit to staying close and it shows up when you have to bring a skeptic along. Telling a client to swap the trade-show booth for a series of small filmed events rarely moves anyone. Going and running two of those events yourself, learning the process, capturing the video, distributing it, and knowing to the dollar what it cost and returned moves everyone. Demonstration outperforms persuasion and you can only demonstrate what you’ve actually done. The leader who delegated everything has nothing to point to but a recommendation.

Strategy is a hypothesis, not a document

The player-coach protects something specific. Strategy isn’t a document you write once and defend; it’s a hypothesis you revise every time the work talks back. The leader who stays in the account hears it talk back constantly in click-through rates and reply rates and the precise reason a channel stopped performing. The leader who delegated all of it hears a summarized version weeks late, already smoothed into a story that flatters the plan. One of them is steering. The other is being driven and calling it leadership.

So the instruction is plain enough to act on this week. Keep one hand on the work no matter how senior the title gets. Not because the grind is noble and not to second-guess your team, but because the work is where strategy comes from, and the day you can’t do it anymore is the day you stop being able to lead it. The operators still standing when the channels shift again will be the ones who never stopped logging in.

Anyone can produce content. It’s table stakes. The harder craft, the one that actually compounds, is producing content at high volume and distributing it so that the right people consume it, share it with influencers and decision-makers inside their own organizations, and begin an internal conversation about the very problem your product solves.

The arrival of generative AI has supercharged this reality. Where producing a single blog post once took hours, AI can now generate a dozen drafts in minutes. This doesn’t solve the marketing problem; it amplifies it. The challenge is no longer the physical act of creation, but the strategic act of differentiation. With the cost of content production plummeting to zero, the value has shifted entirely to the difficult, human-led work: deep audience understanding, genuine insight, and sophisticated distribution. In a world flooded with AI-generated noise, the premium is on content so specific and well-targeted that it can’t be ignored.

The return on all that patient education doesn’t arrive where you spend the effort. It arrives later, as sales efficiency. When buyers finally raise their hands, they come in already qualified. They’re arriving because they already understand your company and what makes it different. They’ve done the reading. They’ve had the internal debate. By the time they reach a salesperson, the hard part of the sale is behind them.

The result is a sales cycle that closes faster, at a higher win rate, with less sales overhead required to produce the same revenue. Fewer reps chasing colder leads. Fewer demos that go nowhere. Fewer deals lost in procurement because the buyer never understood why you were worth a premium in the first place.

Reinvestment is what turns a marketing tactic into a company strategy. That freed-up money, the overhead you no longer need once selling becomes efficient, can be reinvested. You can put it back into marketing, into the product, or into customer success. Fixing the customer education piece is what lets a whole company change shape. The savings don’t just sit on the income statement looking pretty; they fund the next round of advantage.

You Can’t Write for an Audience You Don’t Know

You cannot create the right content until you genuinely understand the audience. Not understand them the way a positioning deck claims to understand them, with three tidy personas and a list of pain points someone invented in a conference room. You have to physically get out into the market. Visit prospects and customers. Go alongside a sales rep, or go alone. Sit in the room, or on the call, and listen to how people actually describe the problem in their own words, with their own anxieties and their own internal politics attached.

What you learn out there is the raw material that feeds the entire content operation. The phrases buyers use, the objections they raise before they raise them out loud, the rival initiatives competing for their budget, the person two desks over who has to be convinced. None of that shows up in a keyword report. You collect it in person, and then everything downstream gets easier because you’re writing from the market instead of writing at it.

What Content Actually Works

Once you understand the audience, a counterintuitive move starts to pay off. Much of your most effective content has nothing directly to do with your product.

This feels wrong to people who measure everything. If we’re paying to publish, shouldn’t every piece sell? No. The most effective content runs parallel to what your audience already cares about. It acts as a gateway. It creates awareness first and earns the right to deliver deeper, product-specific material later. You show up where the buyer’s attention already is, talking about what already keeps them up at night, and you become a voice worth listening to before you’ve asked for anything.

The second lesson is about speed and volume, and most companies fail it badly. They simply don’t produce content fast enough to break through. The reasoning is plain: there’s no real ceiling on volume, because attention across every channel is a matter of supply and demand, and only volume cuts through. You’re competing for finite attention against an effectively infinite supply of other content. One thoughtful post a week loses that math before it starts.

Cutting through also depends on format. Picture your reader honestly. They’re on a phone, on a train, scrolling LinkedIn or Facebook between two other things. They will never wade through a twenty-paragraph essay. So the target is a total read time under two and a half minutes. That constraint isn’t a compromise; it’s the design spec.

A reliable structure for these short posts is to open with industry news and then pivot into your own point of view. The model generalizes to almost any industry. Lead with what just happened in the world your buyer lives in. Then tell them what you think it means. What makes it work is precisely its restraint.

Restraint matters because the action you’re creating is small. It’s a sliver of messaging that quietly begins the process of awareness. You spread the content with paid amplification, advertising spend that pushes a post in front of a targeted audience, yet you’re asking nothing of the reader. No form. No download. No demo. Because you ask for nothing, the buyer’s journey unfolds on its own. Information gets delivered. Conversations start inside the buyer’s organization. Eventually someone comes back to your website, encounters the value proposition (the concise statement of the benefit your product delivers and why it’s worth paying for), and converts into a genuine sales opportunity.

The conventional approach is the classic funnel: a top-of-funnel blog post to attract attention, a mid-funnel asset to download, then a push toward a demo or a price quote. The dashboards look fantastic. But almost none of those leads convert into actual revenue. You’ve optimized for the cost of a hand-raise, not for the readiness of the hand that’s raised.

The conclusion from that failure is the heart of the philosophy. You have to let the buying process happen on its own terms. The marketer’s only job is to deliver the right information at the right time so those internal conversations can occur. You’re not pulling buyers down a funnel. You’re feeding a conversation you don’t control and trusting it to mature.

This is also why generic content fails so badly.

Specificity and the Problem of Volume

Generic content fails because the same vanilla ebook or blog post, blasted out unchanged to everyone, simply disappears into the noise. And the noise is staggering. By a figure that was already a year old, something on the order of two million blog posts were published every single day. That number has likely climbed past two and a half or three million since. Your unremarkable post isn’t competing for attention. It’s drowning in company.

The antidote is specificity. Take one piece of content and tailor it deliberately. Change just a handful of words so that one version speaks to financial advisors and another to investment decision-makers. Same core idea, different reader, different framing of the stakes. These are small, almost trivial tweaks, and most marketers never bother to make them. That’s why they work. The advisor reads a post that sounds like it was written for an advisor, the decision-maker reads one written for a decision-maker, and both feel seen in a stream of content that ignored them. The difference between cutting through and being ignored often comes down to those few words, whether you’re using the content for inbound marketing or for outbound sales outreach.

The specificity problem is really half of a larger neglect. Most people pour ninety percent of their effort into creating content and almost none into figuring out how anyone will actually consume it. Two things have to be true at once. The content has to be genuinely worth consuming, and there has to be a real mechanism to get it in front of people. Either one alone is wasted motion.

Writing a blog post, dropping it on your company website, and having a few employees share it to LinkedIn for twenty views apiece is a waste of time. It feels like marketing. It produces almost nothing. What you need is a high volume of reads that are actually happening, reads by the specific people whose internal conversations you’re trying to start.

Solving that distribution problem is where paid advertising comes back in. Not as a contradiction of the organic philosophy, but as its accelerant. The organic part is what happens after the read: the sharing, the internal debate, the eventual return to your site on the buyer’s own schedule. Paid is simply how you guarantee the read happens at all, in front of the right person, at the volume the supply-and-demand math demands. You’re not buying conversions. You’re buying the first quiet moment of awareness, then getting out of the way.

What This Adds Up To

Done patiently, that machine produces qualified buyers who arrive already convinced, sales cycles that close faster at higher win rates, and overhead you can recover and reinvest. The single piece you fixed turns out to move the whole company. The waste was never the real prize. The real prize was teaching the market to sell itself.

Plenty of founders, especially technical ones, quietly dread selling. The picture in your head is the used-car salesman moving a bad product to someone who doesn’t need it, and you don’t want to be that person. So you try to design selling out of the company: credit-card-only signup, a free trial, anything that avoids getting on a call and pitching. The problem is that scaling revenue, landing stickier customers, and closing bigger deals all run through your ability to sell. No strong salesperson will bet their career on a founder who can’t close.

The good news is that the dread comes from a wrong definition of the job. Get that right and a founder-led process can run above a 50% win rate, and it holds up when you hand it to the salespeople you hire later. Three principles, plus the step most founders skip.

The reframe that makes it bearable

The unlock has nothing to do with closing technique. It’s that selling isn’t convincing someone to buy; it’s getting clear, fast, on whether they should. Once a sales call becomes a diagnosis instead of a pitch, the icky feeling goes away, because you’re no longer talking anyone into anything. You’re a doctor figuring out what’s wrong, and sometimes the honest answer is that this person isn’t your patient. The win rate is a byproduct of disqualifying early, before you’ve sunk hours into the wrong buyer.

Principle one: never go outbound

The first rule is blunt. Don’t do outbound, ever. Get the prospect to come to you. Expect them to fill out a form with real information about their company and role, and book a call on your calendar. Part of the reasoning is timing. Ten years ago a personalized cold email landed because almost nobody was sending them at scale. Today every inbox filters AI-written outreach into the promotions tab, and your prospect already knows an SDR agent wrote it. Outbound is a loser’s game: the response rate sits around 1%.

There’s a corollary worth coaching into every rep: once a meeting lands, you’re allowed to delete it. Look the person up on LinkedIn, check their company, and if they’re outside your ideal customer profile, decline politely and free both calendars. Tell them you want to respect their time and yours, and ask them to correct you if you’ve got it wrong. Protecting reps from bad-fit meetings keeps them working pipeline that actually moves.

And don’t confuse sales and marketing. Getting people to book is a marketing job, turning those bookings into revenue is the sales job, and conflating the two is a common founder mistake.

Principle two: never sell, diagnose

When the prospect shows up, the instinct is to launch into why your product is great. Don’t. The call opens with a question: you could be doing anything right now, so what’s going on in your company that made you book this? Then keep probing. How are you generating pipeline today? What have you already tried? Where are you in the buying process? What specifically are you looking for? None of that is you talking.

Call-analysis tools score how much of a call the rep spoke versus the prospect, and on a strong call the rep is mostly quiet. Drop a transcript into your favorite LLM and ask it to grade the call, and that talk ratio is the first thing it flags. The quality of the questions you ask sets the quality of the call, and the prospect describing their own problem out loud is worth more than any feature you could recite at them.

Principle three: never give a generic demo

The biggest founder mistake isn’t skipping the questions, it’s demoing blind. You’ve got a product with real surface area, every feature maps to some outcome, and people only buy outcomes. So before you share your screen, ask what matters to them. There’s a lot I could show you; what would actually be relevant? If they can’t name a feature, fine, they’ll name an outcome, and you build the demo around that. A walkthrough of everything you’ve shipped is the fastest way to lose a call you’d already half-won.

The rebuttal clever founders reach for is that the product is too new for prospects to know what to ask for. Sometimes true. But even then they’ll tell you the outcome they’re chasing, and an outcome is enough to personalize around. You’re not waiting for them to spec the demo; you’re listening for what they actually want the thing to do.

The step founders skip: define the next step

Here’s the part founders learn the hard way and young ones blow constantly. You run a sharp diagnostic call, you give a tailored demo, and then you close with “cool, I guess I’ll follow up.” That deal is dead.

What you send depends on the deal. A 20-to-30K mid-market deal should close in one call, maybe one and a half, so you send the proposal then. A million-dollar enterprise deal is a six-to-eight-month motion, so the deliverable is a clearly defined next step, not a contract. Either way you budget for it. On a 30-minute call: ten minutes asking why they’re here, ten minutes on questions and a personalized demo, ten minutes defining what happens next, down to who else needs to be in the room.

What you’re actually building

Notice that none of this is advice is closing technique. Come-to-you instead of outbound, diagnose instead of pitch, personalize the demo instead of dumping features, and always name the next step: every principle pulls in the same direction, which is toward telling the truth about fit faster than the other guy. That’s why the dread lifts. You’re not learning to talk people into things; you’re learning to find out, early and on purpose, who you can genuinely help and who you can’t.

That’s also why the playbook survives the handoff. A process built on persuasion lives and dies with the person doing the persuading, so it never transfers. A process built on honest diagnosis is just a set of questions and a discipline, and questions and discipline are teachable. Run it yourself until the win rate proves it works, then hand the salespeople you hire a system they can trust their careers to. Get selling right while it’s still your job, and you build something that keeps selling long after it stops being your job.

Most technical founders carry a quiet ranking in their heads. Research sits at the top. Engineering comes next, a notch below, because engineers implement what the researchers discover. Then, somewhere down in the basement, sits sales and marketing: the dirty work, the part you’d skip if you could. The comforting form of this belief is the line you’ve heard in every Bay Area kitchen: “If we just build a good enough product, it’ll sell itself, and we won’t have to deal with all that sales stuff.”

It’s a delusion, and an expensive one. Call it the Silicon Valley fallacy, and it’s worth naming where it breeds. The idea that a great product sells itself survives because it lets the people who love building things avoid the part of the business they find distasteful. It tells engineers that the work they already enjoy is also the only work that matters. That’s a flattering story. It is also how good products die in obscurity while worse ones take the market.

The fallacy has a cost you can measure. It shows up as a sales team treated as overhead instead of a discipline, hired late, paid on a different floor of respect, and handed a product nobody bothered to make legible to a buyer. It shows up as marketing reduced to “make the website nicer” instead of the work of deciding what category you’re in and what question you want a buyer to be asking when they go looking. Companies that believe products sell themselves tend to build products that require a genius buyer to understand them, then act surprised when the genius buyers are scarce.

Why the lie is so easy to believe right now

There’s a reason this fallacy is having a moment. We’re in a gold rush. Demand for anything with “AI” on it is so hot that founders barely have to sell; buyers are lined up, desperate to sign, eager to throw budget at tokens before they’ve figured out what they want from them. When inbound is that strong, you can run a company with no real sales muscle and post numbers that look like product-market fit. The orders arrive. The dashboard goes up and to the right. Who needs a sales team?

The right picture for this is the astronaut’s. The companies coasting on inbound are like astronauts in zero gravity. Float around long enough with no load on your muscles and they atrophy without you noticing, because nothing is asking them to work. The astronaut feels fine. He feels great, actually; everything is effortless up there. Then he comes back to Earth, gravity switches on, and he can’t stand up.

Gravity always comes back. The AI hangover is already starting: the CIOs who approved six-figure monthly bills are now asking what they bought, the token-maxing debauchery is meeting the morning-after invoice, and the easy inbound will cool the way every gold rush cools. When it does, the companies that built a real GTM discipline while it was unfashionable will still be standing, and the ones who decided their product would sell itself will discover their selling muscles wasted away during the years they didn’t need them. You don’t build a sales organization in the quarter you suddenly need one. You build it before, while you can still pretend you don’t.

“Sales” in this challenge is doing double duty; the phrase is “sales or marketing,” and the marketing half is where the fallacy does its quietest damage. Salesforce is the cleanest case. Marc Benioff is a marketer first, and the “No Software” campaign, the red circle with a slash through the word, decided how a generation of buyers thought about the category before a single rep dialed a number. Gainsight didn’t just sell software; it named “customer success” and convinced a market the function should exist. That naming work is marketing at its most load-bearing, and it’s the part technical founders are quickest to dismiss as fluff. Deciding what category you’re in, what question you want a buyer asking, and what words they’ll use to ask it: that’s not decoration applied after the product is built. It’s a choice that determines whether anyone understands what you made.

The product is the whole journey

The fix isn’t to bolt a sales team onto an engineering culture that quietly looks down on it. That just relocates the contempt. The fix is to redefine what “the product” actually is.

Define the product as the entire journey, from the first time someone hears the company’s name to their tenth renewal a decade later. The software is a large part of that journey. So is the marketing that frames it, the discovery call where a salesperson actually listens, the demo, the solution engineering, the onboarding, the renewal conversation. All of it is the product. Which means the people doing all of it are building the product, and there’s no second class.

The org runs to match. No engineering corner and sales corner. Engineers and salespeople sit intermixed. When a salesperson closes a deal, the engineers say “we closed a deal.” When an engineer ships a feature, the salespeople say “we shipped a feature.” One team, one scoreboard, no priesthood. This is shockingly controversial in the Bay Area, especially in AI. That it’s controversial is the tell. The contempt is so baked in that treating sales as equal reads as a radical act.

You can feel why it works once you stop ranking the functions. The engineer who sees sales as beneath her builds features nobody can sell. The salesperson who sees engineering as a black box sells things the product can’t do. Put them on one team accountable to one business outcome and the engineer starts asking how a customer actually behaves and how to change that behavior so they get more value, and the salesperson starts able to demo what was actually built. The seams between the functions are where deals leak. Sitting people together and giving them a shared number is how you close the seams.

What to do with this if you run GTM

The lesson generalizes past AI startups, because the fallacy isn’t about AI. It’s about respect, and respect shows up in your org chart, your comp, and your calendar.

Hire your first real GTM person earlier than your instincts say, and give the role the same seniority as your first principal engineer. Treating the first salesperson as a hire you’ll “get to once the product is ready” is the fallacy in scheduling form. The product is never ready in the sense that lets it sell itself.

Tie everyone, including engineers, to a business outcome rather than an intermediate metric. “We shipped four features this quarter” is the kind of number that lets a bloated org feel productive while moving nothing that matters. Revenue, retention, the share of a market you actually want: those are the numbers a one-team culture can rally around, and they’re the ones that survive a downturn.

And kill the seating chart that separates the believers from the help. The literal version matters less than the cultural one, but the literal one is a fast diagnostic. If your engineers have never sat in a discovery call and your salespeople can’t get an engineer to return a Slack message, you have two companies pretending to be one, and the gap between them is where your best deals go to die.

The companies that become legendary don’t get there because their product was so good it sold itself. They get there because they refused to believe that was possible, and built the muscle to sell while everyone around them was still floating, weightless, certain they’d never need it.

East of Silicon Valley runs a long set of brown oak-savanna hills that nobody had ever loved. They aren’t redwoods or coastline. The grass stays brown most of the year, and their one real distinction was deeply unglamorous. They happen to be the watershed for the San Francisco Bay, one of the most ecologically critical stretches of ground in the region, roughly the size of Yosemite. The Nature Conservancy (TNC) had wanted to protect them for years and couldn’t get anyone to care. The pitch was a sentence that died on contact: there’s a critical area east of Silicon Valley. Say it out loud. Nothing happens.

So TNC stopped describing the place and named it. They called the brown hills “The Mount Hamilton Wilderness,” after the local peak, and put the name on the map as though it had always been there. The effect was not cosmetic. The Packard Foundation wrote a large grant to protect the Mount Hamilton Wilderness. Local groups around the Bay started campaigning to save the Mount Hamilton Wilderness. Rival organizations began writing the Mount Hamilton Wilderness into their own documents as a settled fact of geography. Mike Sweeney, who ran TNC’s California chapter, would read those documents and laugh because, in his words, “You know we made that up.” Chip and Dan Heath, who tell the story in Made to Stick, land the lesson cleanly: the Mount Hamilton Wilderness is not a set of acres; it’s an eco-celebrity.

This is category creation in its purest form and the most honest example I know, precisely because TNC had nothing to sell. A nonprofit with no product and no profit motive discovered that its single largest strategic unlock was a proper noun it made up. The hills didn’t change and the acreage didn’t change; the word changed, and the word was the part people could finally pick up and carry.

A Name Gives People Something to Carry

People who live in cities know this in their bones. “SoHo” and “the Castro” were never survey coordinates. They’re characters, and the name is what handed each one a personality the street grid never supplied.

A name is a handle on a suitcase. A coherent block of imperiled watershed, packed with ecological argument and competing priorities, is heavy, and most people can’t get their arms around it. Stamp “The Mount Hamilton Wilderness” on the side and anyone can pick it up, point to it, defend it, and write it into a budget. That’s the same work Brian Halligan and Dharmesh Shah did when they drew a circle around blogging, email, and search optimization and called the bundle “Inbound Marketing.” The activities already existed as disconnected chores, and the name turned them into a philosophy you could join, and you joined it by going to HubSpot. It’s what Ward Cunningham did when he looked at tangled code and called it “Technical Debt,” handing engineers a phrase their managers already feared.

Vocabulary Becomes Infrastructure

A coined word does a second thing once it catches: it draws a line between the people who use it and the people who don’t. Learn a tribe’s vocabulary and you’re inside it; every CrossFit box (CrossFit affiliates are called "boxes" rather than "gyms”) runs on private language for exactly that reason. When your customers start saying your word instead of the generic one, leaving you means unlearning it, and that’s a cost most of them won’t pay.

I’ve run this play myself. At Mesosphere (later, D2IQ), while the industry was still fighting over how to run software across thousands of machines at once, we refused to sell “a distributed cluster scheduler.” We named the category the Data Center Operating System and shipped a command-line tool whose command was, literally, dcos. Every engineer who typed dcos node list reinforced the premise in their own hands: the data center is a computer, and a computer runs an operating system. The argument lived in muscle memory instead of in a slide anyone had to be sold on. A name asserts that a thing exists, and steady use quietly turns the assertion into common knowledge.

Name the Outcome

Naming works on funders and buyers by the same mechanism, the one Theodore Levitt described long before any of us got here: people don’t want a four-inch drill, they want a four-inch hole. They back outcomes they can see rather than inputs they can’t. An anonymous run of brown hills is an input. “The Mount Hamilton Wilderness” is the outcome, rendered so vividly you can already feel yourself standing in it. Naming is the craft of making an outcome visible enough that someone wants it badly enough to pay for it.

AI Rewards the Cleanest Handle

The lever has only gotten longer in a market run through AI. Ask a machine to explain your category now, and it reaches for the cleanest handle in everything it has read and repeats that one back to whoever asked. If the handle is the term you coined, used consistently by enough people, you land inside the answer by default, with the credit routing to you whether anyone recalls you started it. The market begins hiring for your word, writing it into job descriptions and analyst notes, and the machine simply mirrors that back. Owning the vocabulary has become the most durable position on the board because it keeps paying out while you sleep.

The Thing Still Has to Be True

One grown-up caution, since it’s the part the hype crowd always skips. None of this is spin. The Mount Hamilton hills genuinely were the Bay’s watershed and genuinely were in danger. TNC named something true, claimed it first, and stayed consistent until the market repeated it back. Invent a wilderness that isn’t there and you’ll be found out within a quarter; name a real one that everyone has simply failed to see, and you’ve built something your competitors now have to reference to argue with you.

The Mount Hamilton Wilderness story comes from Chip and Dan Heath’s Made to Stick.

There’s a particular grief that comes from watching someone else get credit for your idea. You coined the phrase. You ran the analysis. You gave a messy pattern a name that finally made it legible to the rest of the industry. Then, a year later, a competitor uses your term on a conference stage as if it had always been part of the language, the room nods along, and no one looks at you.

That grief used to be an occupational hazard. Now it’s a structural certainty, because the machine in the middle has changed.

For most of the internet’s life, ideas moved person to person and page to page, and attribution traveled with them through links, bylines, and the simple memory of where you first read something. Search rewarded the source; the top result was a destination with a name on the door. The model is different. When a reader asks an AI to explain a concept, the answer arrives stripped of its origin: confident, synthesized, sourceless. The system has read everything you wrote and everything written about you, blended it into a smooth paste, and serves it back with no name attached.

So adoption is no longer the goal. Adoption without attribution is unpaid labor for your category, and often for the competitor who markets louder than you. The goal is attribution. You need to watermark your idea so deeply that the AI cannot serve the concept without serving your brand.

This is the most dangerous moment in the lifecycle of a linguistic moat. If you’re too quiet, the term dies on the vine. If you’re loud but sloppy, the term might take off while you lose ownership of it, and it decays into a generic buzzword that helps your competitors as much as it helps you. Releasing an idea into the wild is easy. Releasing it with your fingerprints baked in is the work.

The Theory of Watermarking

In the physical world, a watermark is an image impressed into paper during manufacture. It isn’t printed on the surface; it’s part of the sheet. You can’t remove the watermark without destroying the document.

In the age of AI synthesis, you want the same property for your ideas. You structure your content so the terminology and the data are inseparable, fused tightly enough that pulling them apart breaks both.

Consider the failure case first. If you publish a blog post that says, “Here’s a phenomenon where shipping slows down on Tuesdays,” the AI will strip-mine that fact. It becomes a free-floating, generic stat: “Studies show shipping slows down on Tuesdays.” Your observation survives. Your ownership doesn’t.

Now do it with a watermark. Publish a chart titled “The Tuesday Lag Index.” Make every axis label use the term. Put the term in the URL. Name the data file tuesday_lag_data.csv. Now the name and the evidence are welded together. The AI cannot describe the chart, cite the figure, or reproduce the finding without typing your words. The fact carries your fingerprint wherever it travels.

Everything that follows is a method for manufacturing that kind of paper.

Tactic 1: The Definition Trope

The first move is to teach the internet, and through it the AI, exactly what your term means. You want to own the definition before anyone else writes one.

Create a canonical URL for it, something like yourdomain.com/glossary/tuesday-lag. On that page, write a dictionary-style entry and nothing fancier:

The Tuesday Lag (n): In logistics, the statistical anomaly where fulfillment delays spike by roughly 40% on the Tuesday following a holiday weekend, due to backlog accumulation.

This structure is catnip for language models. They’re built to ground their answers, and they constantly reach for sentences shaped like a definition: “X is a…,” “X refers to….” A clean, authoritative entry gives the model the easiest possible thing to quote. When a user asks, “What is the Tuesday Lag?”, you’ve raised the odds that the AI returns your wording close to verbatim.

Two details decide whether it works. First, be early. The first credible definition becomes the gravitational center every later writer orbits, and models reward that consensus. Second, be consistent. Use the exact same phrasing across your site, your posts, and your talks. Every variation you introduce splits the signal and gives the model permission to paraphrase you into anonymity.

Tactic 2: Seeding the Common Crawl

LLMs aren’t trained on your marketing brochures. They’re trained on large public scrapes of the web, the Common Crawl chief among them. To get your term into the model’s vocabulary, you plant it where the crawl reaches and where the training process assigns weight.

1. Reddit and forums. Start threads that discuss the phenomenon by name. “Has anyone else hit The Tuesday Lag after this last long weekend?” Models lean on Reddit because it reads as real people reaching real consensus, which is exactly the texture you want your term to acquire.

2. Wikipedia, the holy grail. If the concept is rigorous enough to survive editorial scrutiny and is backed by third-party citations, a Wikipedia entry is the decisive win. It functions as long-term memory for the AI, and it lends your term an air of settled fact.

3. YouTube transcripts. Spoken-word content increasingly feeds the training data. Record yourself explaining the concept, say the term plainly and often, and the auto-generated transcript seeds it in a format the crawl ingests cleanly.

The connective thread across all three is repetition of the exact string. A term mentioned once in a hundred clever paraphrases barely registers. The same words, repeated in many independent places by many different people, is the pattern a model learns. You’re not trying to be original each time. You’re trying to be consistent enough that the phrase calcifies.

Tactic 3: The Trojan Horse Asset

One of the most durable ways to watermark an idea is to bury it inside something useful, so people carry it for you without thinking about it.

In the Kubernetes war, we didn’t just write blog posts about the “Data Center Operating System.” We built a command-line interface where the command itself was dcos:

• dcos package install spark

• dcos node list

Every time an engineer typed that command, they reinforced the premise that this thing was an operating system. The argument wasn’t in a whitepaper anyone had to be persuaded by. It was in their muscle memory.

You might not ship a CLI, but the shape generalizes:

• A calculator. “Calculate your Tuesday Lag Risk Score.”

• A template. “The Tuesday Lag Staffing Adjuster.”

• A certification. “Certified Inbound Marketer.”

When a utility is genuinely valuable, people share it, and the name is welded to the tool. They can’t pass along the thing that helps them without passing along your vocabulary. Usage becomes repetition, repetition becomes training signal, and the watermark spreads on the strength of the favor it does for everyone who touches it.

Defensive Positioning: When They Steal It

If any of this works, competitors will start using your term. Your instinct will be to panic, to reach for a cease-and-desist. Don’t.

When a competitor adopts your terminology, they’re capitulating. They’re conceding that your map of the territory is the correct one, and they’re agreeing to argue on the ground you surveyed.

When Microsoft began talking about “Inbound Marketing,” HubSpot didn’t sue. They celebrated. If Microsoft is talking about Inbound, and HubSpot is the first result anyone finds for Inbound, then Microsoft is effectively buying ads for HubSpot. The theft routes attention straight back to the source.

The moment your competitors adopt your language, the game is over and you’ve won it. You’ve stopped being one vendor among several and become the standard everyone else has to reference.

The Ultimate Metric: Citation Velocity

In the old world, we measured keyword volume: how many people typed a phrase into a search box. In the new world, measure citation velocity, the rate at which your term shows up in the language other people use on their own.

How often does it appear in industry newsletters you don’t write? In analyst notes? In job descriptions? “Looking for a manager familiar with Inbound methodologies” is worth more than a thousand impressions, because a company has rewritten its own hiring around your word. That’s the market speaking your language back to itself, unprompted.

This is the measure that matters, because the AI is essentially a mirror of the market. When enough of the market describes its own reality in your terms, the model reflects that reality back to every person who asks it a question. The credit you once had to fight for becomes the path of least resistance for the machine.

That’s the whole point of the watermark. The grief at the start of this essay, watching your idea walk off without you, came from attribution being something fragile that you had to defend. Build the watermark well and that fragility becomes your defense: the more your idea spreads, the harder your name is to remove from it. You’re no longer a source someone has to remember to credit. You’re part of the answer itself.

If you walk into a marketing agency today and ask them to increase your lead flow, they will almost certainly talk to you about “Inbound Marketing.”

They will use the term casually, as if it were a fundamental force of nature like gravity or electromagnetism. But “Inbound Marketing” is not a natural phenomenon. It was invented in a conference room in Cambridge, Massachusetts, by the founders of HubSpot, Brian Halligan and Dharmesh Shah.

Before HubSpot, the activities that make up Inbound Marketing already existed. Companies were already writing blogs. They were already sending emails. They were already optimizing their websites for search engines. But these activities were disparate, disconnected chores. They were just “stuff marketing interns did.”

HubSpot’s genius wasn’t just building software to manage these tasks. Their genius was grouping these messy, disconnected activities together and giving them a name.

By coining the term “Inbound Marketing,” they created a conceptual container. They drew a circle around a specific set of behaviors (blogging, SEO, social media) and labeled it “Inbound.” Simultaneously, they labeled everything else, cold calling, TV ads, spam, as “Outbound.”

This wasn’t just a branding exercise; it was a coup.

Suddenly, you weren’t just “blogging”; you were practicing a philosophy. You were part of a movement. And crucially, if you were eager to learn how to do this new thing, you had to go to the people who named it. You had to go to HubSpot.

The Handle on the Suitcase

A simple label changes the economic value of an activity, and the reason sits in cognitive science. The human brain is a cognitive miser, constantly looking for ways to conserve energy. Processing complex, nuanced, or disconnected information is metabolically expensive.

Imagine a complex methodology, like the Kubernetes orchestration from the last post or HubSpot’s marketing philosophy, as a heavy, awkward suitcase. It’s packed with data, protocols, and exceptions. Without a handle, it’s almost impossible to pick up. You have to wrap your arms around it, heave it up, and shuffle forward. It’s exhausting.

A specific term acts as the handle on the suitcase.

When you coin a term like “Inbound Marketing” or “Data Center Operating System,” you are compressing a massive amount of information into a portable linguistic unit. You are giving the brain a shortcut.

Now, instead of explaining the entire history of container orchestration every time we want to discuss it, we just say “DCOS.” The brain accesses the label, unpacks the associated meaning, and moves on.

This is why “generic” content fails in the AI era. Generic content describes the contents of the suitcase (“Here is a sock, here is a shirt, here is a toothbrush”). It’s accurate, but it’s heavy. Proprietary terminology gives the user the handle. And the AI, just like the human brain, prefers the handle because it is the most efficient way to retrieve the concept.

Affordances for the Mind

In design theory, there is a concept called an “affordance.” A flat plate on a door affords pushing; a vertical bar affords pulling. The physical shape of the object tells you how to interact with it.

Words are cognitive affordances. They tell us how to interact with an idea.

If you don’t have a word for a problem, you can’t solve it. You can’t even really discuss it.

Consider the term “Technical Debt.” Before Ward Cunningham coined this metaphor, software engineers struggled to explain to management why they needed to refactor code. They would talk about “messy code” or “spaghetti logic,” which sounded like laziness.

But when Cunningham slapped the label “Debt” on it, he created a cognitive affordance. Managers understand debt. They know that if you don’t pay down the principal, the interest accumulates until it bankrupts you. Suddenly, the problem had a shape. It had a handle. Managers could “pick up” the concept and make decisions about it.

In your industry, there are dozens of problems that your customers are facing right now that they cannot solve because they cannot name. They are struggling with a vague sense of inefficiency or a friction they can’t articulate.

If you’re the one who hands them the language to describe their pain, the one who gives them the handle, they’ll trust you to solve it.

The Velvet Rope: Tribalism and Insider Status

Finally, proprietary terminology works because humans are tribal creatures. We use language to signal who is “in” and who is “out.”

If you walk into a CrossFit gym, you will hear a foreign language: WOD, AMRAP, Rx, Box, Burpee. If you know these words, you are part of the tribe. If you don’t, you are a tourist.

When you create a unique taxonomy for your business, you are inviting your customers into a tribe. You are saying, “We see the world differently here. We have our own words for things.”

This is incredibly sticky. Once a customer learns your language, once they start calling it “Inbound” instead of “digital marketing,” or “The Tuesday Lag” instead of “shipping delays,” switching costs skyrocket. To switch to a competitor, they would have to unlearn your language and learn a new one. They would have to give up their insider status.

The AI Connection

The same logic governs AI, which turns the challenge into an opportunity.

AI models are the ultimate outsiders desperately trying to become insiders. They are trained to mimic human language patterns, to blend in, to sound like they belong. When an AI encounters a term like “Inbound Marketing” or “Technical Debt” being used consistently by a tribe of experts, it assigns a high statistical weight to that term. It recognizes it as a “handle” for a specific cluster of concepts: a shortcut to efficiency.

The AI, like the human brain, is a cognitive miser. It prefers the handle because it’s the most efficient way to retrieve and package the concept.

If you successfully build a tribe around your terminology, if the market begins using your language consistently, the AI has no choice but to adopt it. It cannot accurately answer a question about modern marketing without using the word “Inbound,” because the training data overwhelmingly signals that “Inbound” is the correct handle for that conceptual suitcase.

And because you coined the term, you own the definition.

This is the fundamental shift from the SEO era to the AI era. In the old world, you won by ranking #1 for “marketing strategies.” In the new world, you win by making it impossible for anyone, human or machine, to discuss marketing strategies without using your vocabulary.

You’ve forced the AI to play by your rules, on your field.

There is a story first-time founders tell themselves, usually without realizing they’re telling it. It goes like this: I will build the product, I will launch the product, people will discover the product, and the company will begin. The story has a satisfying shape. It has a beginning, a middle, and a launch day. It is also, in almost every case, fiction.

What actually happens is stranger and slower. You build the product. You launch it. A modest number of people look at it, a smaller number try it, and a much smaller number stick around. The metrics don’t say “yes” and they don’t say “no.” They say something closer to “hmm.” And then you enter a phase of company-building that nobody adequately warned you about: a long, ambiguous, often demoralizing period of adjusting what you’ve built and how you talk about it, over and over, until something clicks. Or doesn’t.

This essay is about that period. It’s about two ideas that sound like jargon but describe the most important work you will do as a founder: product/market fit and message/market fit. If you’ve never worked in go-to-market roles before, which is to say sales, marketing, or anything else concerned with how a company finds and wins customers, these terms may be new to you. That’s fine. The concepts are simple to define and brutally hard to achieve, and the gap between those two facts is where most startups die.

What product/market fit actually means

The term “product/market fit” was popularized by Marc Andreessen in a 2007 blog post, building on thinking from his Stanford professor and benchmark investor Andy Rachleff. Andreessen’s definition is almost anticlimactic: product/market fit means being in a good market with a product that can satisfy that market.

Let’s unpack that, because every word is doing work.

A “market” is not “everyone who might conceivably use this.” A market is a group of people or companies who share a problem, know they have it, and are willing to spend money or change their behavior to solve it. That last clause matters most. Plenty of problems are real but not urgent. People will agree, in conversation, that yes, that’s annoying, that’s inefficient, somebody should fix that. Then they will go back to their day and never think about it again. A real market is made of people for whom the problem is painful enough that they are already trying to solve it, badly, with spreadsheets and duct tape and workarounds. The existence of ugly workarounds is one of the most reliable signals of a real market because it proves people care enough to suffer for a solution.

A product that “satisfies” that market is one that solves the problem well enough that people adopt it, keep using it, and would be upset if you took it away. Not impressed by it. Not complimentary about it. Dependent on it.

Product/market fit, then, is the state where these two things connect: a genuine, motivated demand and a product that genuinely answers it. When you have it, you can feel it. Customers pull the product out of your hands. Word of mouth starts happening without you. Your problem shifts from “how do we get anyone to care” to “how do we keep up?” Rachleff liked to say that you can always feel when product/market fit isn’t happening, and you can always feel when it is, and the difference is unmistakable.

The corollary, which founders resist, is that if you’re not sure whether you have product/market fit, you don’t have it.

What message/market fit means, and why nobody tells you about it

Product/market fit gets all the attention. Message/market fit is its quieter sibling, and ignoring it is one of the most common ways technical founders sabotage perfectly good products.

Message/market fit means you have found the words. Specifically: you have found a way to describe what your product does, for whom, and why it matters, such that the right people hear it and immediately understand that it’s for them. The message lands. It doesn’t require explanation or a demo or a twenty-minute call to make sense. The prospect reads your home page or hears your one-liner, and something in their head goes: oh. That’s my problem. That’s me.

Here’s why this matters as much as the product itself: your product cannot speak for itself. This is one of the harder truths of early-stage company building. You believe, deeply, that if people would just try the product, they would get it. You’re probably right. But “just try the product” is itself a sale, and the only tool you have to make that sale is language. Before anyone experiences what you built, they experience how you describe what you built. If the description fails, the product never gets its chance. A great product with a bad message loses to a decent product with a great message far more often than engineers want to believe.

And here is the part that surprises first-time founders most: you, the person who understands the product best, are usually the worst-positioned person to find the message. You know too much. You describe the product in terms of what it is and how it works because that’s how you think about it, having built it. The customer doesn’t care what it is. The customer cares what it does for them, in their language, against the backdrop of their existing frustrations. The message that works is almost never the message you would have written on day one, and you cannot derive it from first principles. You have to find it the same way you find product/market fit: by trying things in front of real people and watching what happens.

Two searches, one loop

It would be convenient if these were sequential, if you could nail the product first and then figure out the messaging. They’re not, and you can’t, because each one is the instrument you use to measure the other.

Think about what happens when you put your product in front of a prospect and they don’t bite. What went wrong? Maybe the product doesn’t solve a problem they care about. Maybe it solves a problem they care about, but your description of it was so off-key that they never realized it. Maybe the product and message are both fine and this just wasn’t the right person. From the outside, all three failures look identical: a shrug, a polite “interesting,” a trial that goes nowhere. The signal doesn’t come labeled.

This is why early-stage iteration is so disorienting. You are running an experiment with at least three free variables, the product, the message, and the audience, and every result is ambiguous about which variable caused it. A scientist would call this a confounded experiment and redesign it. A founder doesn’t have that luxury. You can’t isolate variables when each test takes weeks and your runway is measured in months. So you do the unscientific thing: you form a judgment about which variable is most suspect, change it, and run the loop again.

The loop itself is simple to describe. You make a hypothesis: these people have this problem, our product solves it, and these words will make that obvious. You expose the hypothesis to reality through sales calls, a landing page, cold outreach, a demo, and a launch. You collect the signal, which is usually faint and contaminated with noise. You revise the hypothesis. You go again.

What nobody tells you is how many times you’ll run that loop. Not three. Not five. Often dozens, across a year or two, with most iterations producing results that are slightly worse than ambiguous. The companies whose origin stories you’ve read have compressed this period into a sentence, usually something like “after some early experimentation, we found our footing.” That sentence is hiding eighteen months of someone’s life.

The signals that lie to you

Part of what makes the search slow is that the early evidence actively misleads you in both directions. It’s worth cataloging the major lies because you will encounter every one of them.

The first lie is enthusiasm without commitment. People are nice. When you show someone your product, especially someone who knows you, they will say encouraging things. “This is really cool.” “I could see a lot of people using this.” “Send me a link when it launches.” None of these statements cost the speaker anything, and statements that cost nothing carry no information. The only signals that matter are the ones with a price attached: money, time, reputation, switching cost. Did they pay? Did they integrate it into their workflow? Did they introduce you to a colleague, putting their own credibility behind you? Did they come back next week without being prompted? Compliments are noise. Behavior is a signal.

The second lie is the wrong customers saying yes. Sometimes you’ll get traction, real, paying traction, from people outside the market you intended to serve. This feels like success and might be. But it might also be a trap: a small pocket of demand that doesn’t generalize, pulling your roadmap toward customizations that serve five customers and repel five thousand. Early money is intoxicating precisely because you have so little of it, and it takes discipline to ask whether the people buying represent a market or merely a handful of anomalies.

The third lie is the false negative. You pitch the product to twenty people, nobody converts, and you conclude the product is wrong. But maybe the product was right and the message was wrong, or the message was right and those twenty people were the wrong audience. Promising companies have been abandoned over results that condemned the words, not the work. This is why the entanglement of product fit and message fit is so dangerous: a failure in either one can masquerade as a failure of the whole.

The fourth lie is the metric mirage. Signups that don’t activate. Traffic that doesn’t convert. A launch-day spike from a community that came for the novelty and never returned. Vanity metrics are called that because they flatter you, and flattery, again, is noise. The metrics that matter in the search phase are unglamorous: retention, repeat usage, the percentage of trials that turn into habits. They move slowly, they’re small, and they tell the truth.

What progress actually looks like

Given all this noise, how do you know you’re getting anywhere? The honest answer is that progress in the search phase rarely looks like progress. It looks like a slow accumulation of small, qualitative shifts, most of which you’ll only recognize in hindsight. But there are things to watch for.

The first is a change in the texture of conversations. Early on, you do all the talking. You explain, you demo, you handle objections, and the prospect mostly nods. Somewhere along the way, if things are working, the balance shifts. Prospects start finishing your sentences. They describe their problem using language eerily close to your positioning, sometimes before you’ve said it. They ask questions that presume they’re going to use the product (“how does this handle our SSO setup?”) rather than questions that audit whether they should (“so who else is using this?”). When the conversation starts pulling instead of needing to be pushed, something has changed.

The second is the boomerang. Someone you talked to months ago, who went quiet, comes back. Or someone you’ve never met shows up having heard about you from a customer. Unprompted return and unprompted referral are among the strongest signals in early-stage life because nobody fakes them out of politeness.

The third is repetition in the feedback. In the early days, every customer conversation surfaces different complaints and different feature requests, which is its own kind of signal: it means you haven’t found a coherent market yet because a coherent market wants coherent things. When the feedback starts converging, when the fifth and sixth and seventh customers all ask for the same thing and describe the same pain, you’ve located something real. Even if the convergent feedback is negative, it’s a gift, because at last the experiment is producing legible results.

The fourth, specific to message/market fit, is theft. Customers start using your words. They describe your product to their colleagues using the framing you developed, sometimes verbatim. When your language escapes your own marketing and starts circulating on its own, the message has fit the market. Until then, every description a customer gives of your product is data: the words they choose, the comparisons they reach for, the parts they emphasize and the parts they leave out. Their language is usually better than yours. Steal it back.

How to conduct the search without losing your mind

There’s no formula for this, and you should be suspicious of anyone selling one. But there are postures that make the search faster and saner, and they’re worth naming.

Write your hypotheses down. This sounds bureaucratic and is actually liberating. Before each iteration, state in plain language what you currently believe: who the customer is, what their problem is, why your product wins, and what words you’re betting will land. The point isn’t formality. The point is that an unwritten hypothesis mutates silently to fit whatever happened, and then you learn nothing. A written hypothesis can be wrong, and being wrong on paper is how you make progress. Six months in, the document also becomes a record of how far your thinking has traveled, which on the bad days is worth more than you’d expect.

Change fewer things at once. You can’t run controlled experiments, but you can avoid maximally confounding yourself. If you rewrite the home page, reposition the product, change the pricing, and target a new segment all in the same month, whatever happens next will teach you nothing, because you won’t know what caused it. Sequence your changes when you can. Hold the product steady while you test a new message; hold the message steady while you test a new audience.

Spend more time listening than building. This is the hardest discipline for technical founders, because building feels like progress and conversations feel like a distraction from progress. It’s exactly backwards. In the search phase, the scarce resource isn’t code, it’s information about why people do and don’t buy, and that information lives exclusively in the heads of your prospects and customers. Every week that passes without direct customer conversation is a week of flying blind. And when you do talk to customers, resist the urge to pitch. Ask about their problem, their current workaround, and the last time the pain flared up. The questions that begin “would you use a product that...” are nearly worthless because people are terrible at predicting their own behavior and excellent at being agreeable. Ask about what they’ve already done, not what they might do.

Define, in advance, what would change your mind. Persistence is a founder virtue right up until it becomes denial, and the line between the two is invisible from the inside. The protection is to decide, while you’re still clear-headed, what evidence would convince you that a hypothesis is dead: if we can’t get ten paying customers from this segment in ninety days, the segment is wrong. You won’t always honor these tripwires, but having them at least forces a real conversation when you cross one.

And calibrate your expectations about time. The honest range, for most companies, is one to two years of searching before product/market fit, sometimes longer, and the search for message/market fit continues in some form forever, because markets shift and language wears out. If you budget six months for this phase, you will spend month seven onward feeling like a failure for what is, in fact, the normal pace of the work.

You are not executing yet. You are searching.

Here, finally, is the thing I most wish someone had told me to tell first-time founders, because it changes how the whole frustrating period feels.

Most of what you’ve absorbed about running a company, from business books, from press coverage, from watching big companies operate, is about execution: setting goals, hitting numbers, scaling what works. Execution assumes you know what works. In the pre-fit phase, you don’t, and pretending otherwise is the root error behind most early-stage flailing. Founders set growth targets before they know what they’re growing. They hire salespeople before they know what sells, marketers before they know what message lands. They scale the part of the company that’s supposed to amplify a signal before the signal exists. Steve Blank built an entire methodology around this distinction: a startup, in his definition, is not a small version of a big company but a temporary organization searching for a repeatable business model. The search and the execution are different jobs requiring different mindsets, and the failure to notice which job you’re doing is fatal more often than any competitor is.

Once you accept that you’re searching, the frustration recontextualizes. The ambiguous signals, the discarded positioning, the pivot that erased three months of work: these aren’t evidence that you’re failing at execution. They’re the ordinary texture of search. A failed experiment that kills a bad hypothesis is progress, even though it feels like loss. The landing page that converted nobody told you something true. The segment that wouldn’t pay narrowed the map. Search progress is measured in eliminated possibilities and sharpened hypotheses, not in revenue, and if you measure this phase by execution metrics you will conclude, falsely and miserably, that nothing is happening.

Something is happening. It’s just slow, and it’s supposed to be. Product/market fit and message/market fit are not milestones you schedule; they’re discoveries you earn, one ambiguous conversation at a time, by staying curious slightly longer than it’s comfortable to. The founders who get there are rarely the ones with the best initial idea. They’re the ones who treated their idea as a draft, their words as experiments, and their customers as the only authority that matters, and who kept running the loop after the romance of launch day wore off.

The story you should tell yourself isn’t “build it, launch it, grow it.” It’s “guess, test, listen, revise.” Less cinematic, admittedly. But it has the advantage of being how it actually works.

In 2014, the world of cloud infrastructure was a chaotic mess. Docker had just exploded onto the scene, popularizing “containers,” a technology that allowed developers to package code in neat, portable boxes. Everyone knew containers were the future, but nobody knew how to manage them at scale.

If you had one container, you were fine. If you had ten, you could manage them manually. But if you had a thousand containers running across a hundred servers, you were in hell.

This was the “Orchestration War.” The combatants were heavyweights. There was Docker Swarm, backed by the company that invented the container craze. There was Kubernetes, an open-source juggernaut birthed from the belly of Google. And then there was us: Mesosphere.

The Pedigree Trap

To understand the intensity of this war, you have to understand the pedigree of the technology we were bringing to the fight. Apache Mesos wasn’t just a random open-source project; it was born in the legendary AMP Lab at UC Berkeley.

The AMP Lab was an innovation factory. It was where Matei Zaharia created Apache Spark, which was actually the very first application built to run on Mesos. Matei went on to found Databricks, and his peer, Ben Hindman, founded Mesosphere.

We were banking on a historical pattern that everyone in Silicon Valley recognized. Years earlier, Google had built a proprietary internal file system and a processing framework called MapReduce. They wrote academic papers about them but kept the code secret. Engineers at Yahoo took those papers and built an open-source clone called Hadoop (HDFS). Companies like Cloudera then built massive businesses monetizing that open-source version.

We thought history was repeating itself. Google had built a secret internal system called “Borg” to manage their massive data centers. They wrote a paper about it. Ben Hindman and the AMP Lab students took the concepts from that paper and built Mesos. The logic was clear: Google writes the paper, we build the open-source version, and then we build the big company.

But this time, Google broke the pattern. They didn’t just release a paper. They released code. They launched Kubernetes, an open-source version of Borg, directly into the market.

Suddenly, we weren’t just competing against another startup. We were competing against the source. We couldn’t out-Google Google.

The 40,000-Core Laptop

If we tried to compete on “features”—who has better networking, who has better scheduling—we would lose. So, we decided to change the battlefield. We decided to stop selling container orchestration software and start selling a new history of computing.

Our linguistic strategy began with a question, one we repeated in every meeting, every keynote, and every press briefing until we were sick of the sound of our own voices:

“Why are the 40,000 cores in your data center treated any differently than the 4 cores in your laptop?”

It was a simple, brutal question. It highlighted the absurdity of the status quo. When you open your laptop to run an application, you don’t manually tell the processor which core to use. You don’t allocate RAM by hand. You don’t wire the networking cables. The Operating System (OS) does that for you. The OS is the magic layer that abstracts the messy hardware into a clean, usable interface.

But in the data center? In 2014, engineers were doing exactly that—manually pinning tasks to servers, hard-coding IP addresses, and suffering through what we affectionately called “brain damage.”

The data center was a computer without an operating system.

Naming the Category

Once we identified the gap, we didn’t just say, “Our software automates this.” We named the solution. We called our product the Data Center Operating System (DC/OS).

This wasn’t just a tagline; it was a taxonomy. By using the term “Operating System,” we instantly inherited fifty years of mental models. Everyone knows what an OS is. Everyone knows you need one. You wouldn’t buy a laptop without Windows or macOS. So, why would you build a data center without a DC/OS?

We created a visual narrative—a slide that became famous in the industry. It mapped the eras of computing not just by hardware, but by the dominant operating system that defined them:

1. The Mainframe Era (One computer, many users) — Dominated by IBM’s OS/360.

2. The PC Era (One computer, one user) — Defined by DOS and later Windows.

3. The Server Era (One computer, one to many users) — Ruled by Linux and Windows Server.

4. The Mobile Era (One computer, mobile user) — The age of Android and iOS.

5. The Data Center Era (Many computers, one system) — ?

For that last era, the slot was empty. There was no operating system for the data center—until DC/OS.

We used this gap to drive a powerful PR narrative. We told reporters, “This is a historic moment. This is the first new operating system released in a decade. It’s the first operating system released since Android.”

The Litmus Test: Real vs. Fake

We didn’t just stop at the high-level metaphor. We took the terminology deep into the technical weeds. We knew that if we stayed vague, competitors could wave their hands and say, “We do that too.” So, we published a strict rubric: “How to tell a real Operating System from a fake one.”

We argued that to legally call yourself an operating system, you had to perform specific duties. We listed the core requirements of any OS, from the mainframe to the iPhone, and applied them to the data center:

1. Spawn and isolate tasks: You must be able to start processes and keep them from crashing each other.

2. Manage memory: You must allocate RAM dynamically across the fleet.

3. Have a file system: You need a way to store and retrieve data abstractly.

4. Manage networking and I/O: You must control the flow of data in and out.

5. Have an API: Developers need a programmatic way to talk to the kernel.

6. Have a CLI: Operators need a command-line interface for scripting and control.

7. Have a Graphical UI: Humans need a visual abstraction to understand the system state.

This list was a trap. By defining these criteria based on the gold standard of Linux and Windows, we created a checklist that only we could pass. If a competitor didn’t have a GUI? “Sorry, you’re just a kernel, not an OS.” If you didn’t handle storage? “You’re just a scheduler.”

By manipulating this language, we effectively “locked out” other systems. They simply weren’t prepared to respond to a framework that defined the battlefield so narrowly. And this created a strategic catch-22 for them: if they ignored our definition, they looked incomplete. But even if they did respond—arguing that a GUI wasn’t necessary, or that they had a different approach—they still lost. Why? Because by arguing with us, they were acknowledging our framework. They were fighting on our turf, validating that we owned the definitions.

Changing the Physics of the Market

The effect of this linguistic pivot was immediate and violent.

By framing the market as “Who will build the Data Center OS?”, we forced our competitors to fight on our terrain. Kubernetes wasn’t positioned as an operating system; it was a “scheduler.” Docker Swarm was a “clustering tool.” Compared to an “Operating System,” those sounded like features.

Suddenly, analysts at Gartner and Forrester weren’t asking, “Does Mesosphere support pod autoscaling?” They were asking, “Is Kubernetes a true Data Center OS?”

We had successfully watermarked the market with our terminology. For the first year of that war, we beat the pants off Kubernetes in terms of share of voice and enterprise attention. We landed massive contracts with Fortune 500 companies not because our feature list was longer, but because our story was clearer. We sold the C-suite on the idea of an Operating System for their cloud, while our competitors were down in the weeds selling APIs to engineers.

The Lesson: The Definition is the Defensibility

Now, the history books will tell you that Kubernetes eventually won the war. And they did. Mesosphere made critical business errors—we kept key parts of our stack proprietary for too long while Kubernetes was open and free. The power of “free” combined with Google’s engineering might eventually overwhelmed our linguistic moat.

But that doesn’t negate the lesson; it amplifies it.

For eighteen months, a smaller startup held off the combined might of Google and Docker entirely through the power of language. We didn’t have more engineers. We didn’t have more money. We had a better taxonomy.

In the age of AI, this lesson is your survival guide. If you allow the market (or the LLM) to define the category, you are a commodity. If you define the category yourself—if you name the pain and name the solution—you become the standard.

One of the most common tools in the SEO toolbox was known as the “Skyscraper Technique.”

The premise was logical, almost scientific. If you wanted to rank for a competitive term like “Project Management Best Practices,” you would look at the current number one result on Google. If that article had ten tips, you would write an article with twenty tips. If their article was 1,500 words, yours would be 3,000. You would add better graphics, more current data, and a catchier headline. You would build a taller building, and Google, recognizing your superiority, would crown you the winner.

This strategy worked because Google was a sorting engine. Its job was to rank discrete documents. It compared Document A to Document B and decided which one was “better.”

But in the age of synthesis, the Skyscraper Technique is ineffective.

When an LLM answers a user’s prompt, it is not looking for the best document. It is looking for the consensus answer. It is scanning the skyscrapers, the two-story houses, and the shacks, and it is calculating the statistical average of what “Project Management” looks like.

If you write a 5,000-word definitive guide that perfectly mirrors the consensus view of your industry, you haven’t built a moat. You have simply contributed a very high-quality brick to the wall of noise that the AI uses to train itself.

The Mechanism of Mediocrity

To understand why “better” doesn’t beat AI, you have to respect the underlying math of the Large Language Model.

At their core, these models are prediction engines designed to generate the most probable next token (word) in a sequence. They are trained on the internet, which means they are trained on the average of human expression. When you ask an AI a question, it is statistically incentivized to give you the most likely answer, the answer that aligns with the majority of its training data.

This means LLMs are, by definition, engines of conformity. They smooth out the edges. They remove the outliers. They regress to the mean.

If you are a content creator following traditional SEO advice, you are likely using tools like SEMrush or Ahrefs to find “keywords with high search volume.” You see that 50,000 people a month search for “How to write a cover letter.” So, you write a guide on cover letters. You look at what’s already ranking and you make sure to cover the same points, because that’s what Google expects.

By doing this, you are explicitly training the AI to ignore you. You are telling the model, “I agree with everyone else.”

When the AI synthesizes an answer for the user, it will output the generic, consensus advice. It won’t cite you because you didn’t say anything unique. You were just part of the choir.

The Canary in the Kitchen: The Recipe Apocalypse

If you want to see the future of your industry, look at what happened to recipe blogs.

For years, recipe bloggers were the masters of the SEO game. They knew that to rank for “Chocolate Chip Cookies,” they couldn’t just post a recipe. Google required “dwell time” and “content depth.” So, bloggers wrapped their ingredients in 2,000-word narratives about their grandmother’s farmhouse, the smell of autumn, and the emotional resonance of melted chocolate.

Users hated it. They mocked the “life story before the recipe.” But it worked. It drove traffic, which drove ad impressions, which drove revenue.

Then came the AI.

If you ask ChatGPT for a “chewy chocolate chip cookie recipe,” it does not give you a story about autumn. It gives you the ingredients and the instructions. It extracts the data (the recipe) and discards the wrapper (the narrative).

The value of the blog post was the data. The wrapper was just friction. The AI removed the friction, and in doing so, it destroyed the business model. The user got the cookie, and the blogger got zero clicks.

This is the Great Flattening. The AI separates the insight from the creator. If your insight is generic (flour, sugar, butter), you are anonymous.

The Coder’s Dilemma

The developer community saw this same dynamic unfold in real time, and the numbers tell a brutal story.

For over a decade, Stack Overflow was the oracle of programming knowledge. If you had a coding problem—a cryptic error message, a tricky algorithm, a framework you couldn’t quite grasp—you had a ritual: search Google, land on a Stack Overflow thread, scan for the green checkmark, copy the solution.

This ecosystem thrived on a simple exchange. Developers contributed their expertise for free, building reputation through upvotes and accepted answers. In return, Stack Overflow captured massive traffic, which it monetized through ads, job listings, and enterprise products. The contributors got status. The platform got revenue. The searchers got solutions.

Then came GitHub Copilot and ChatGPT.

Now, when a developer needs to write a function, they don’t search. They don’t even leave their IDE. They simply describe what they want—“write a Python function to reverse a string”—and the AI generates the code instantly. The AI, having been trained on millions of Stack Overflow threads, synthesizes the consensus solution without attribution, without a visit, without a click.

Traffic to Stack Overflow didn’t just decline. It collapsed. Between 2022 and 2024, the site saw a 50% drop in new questions posted and a corresponding nosedive in page views. The contributors who had spent years building their reputation scores—the “10k rep” users who were the platform’s aristocracy—suddenly found themselves ghostwriting for machines that would never credit them.

Why? Because the “answer” was a commodity.

The code to reverse a string in Python is always the same. It doesn’t matter who wrote it first, who explained it best, or who has 50,000 reputation points. The AI doesn’t care about expertise signals. It cares about statistical patterns. It reads every solution, averages them out, and outputs the consensus.

Stack Overflow was a cathedral built on consensus knowledge. And consensus is undoubtedly what AI consumes without attribution.

The contributors weren’t providing unique insight. They were documenting standard practice. They were building the corpus that would make their own platform obsolete. Every well-formatted answer, every carefully explained algorithm, every “here’s how you do this in Python 3.8” was training data for the model that would eventually replace the need to visit Stack Overflow at all.

This is the paradox of commodity expertise: The better you document the consensus, the more efficiently the AI can extract it.

And here’s the kicker—Stack Overflow’s model was designed to create consensus. The voting system promoted the most popular answers to the top. Duplicate questions were closed and redirected. Unconventional solutions were downvoted as “not idiomatic.” The entire platform was a consensus-generating machine.

Which made it the perfect training ground for an AI that specializes in outputting the most statistically likely answer.

The Trap of High Volume

This leads to a terrifying conclusion for modern marketing: Targeting high-volume keywords is now a liability.

In the SEO era, high volume was the prize. It meant a large addressable market. It signaled commercial intent and justified budget allocation. You could point to a keyword getting 50,000 monthly searches and make a compelling business case for creating content around it.

But in the AI era, high volume implies high consensus. If everyone is searching for it and everyone is writing about it, then the AI has already mastered it. The corpus is complete. The statistical average is locked in.

If you try to compete for “CRM software,” you are fighting a losing battle against a machine that has read every review of Salesforce and HubSpot ever written. The AI can synthesize a comparison table in seconds that is more objective, more comprehensive, and more up-to-date than your blog post could ever be.

You cannot win by answering the questions that everyone is asking. You cannot win by providing the answers that everyone expects.

This is the Great Flattening in action. The more common the question, the more commoditized the answer. The more traffic a keyword drives today, the less value it will have tomorrow.

The SEO playbook told you to climb the mountain toward consensus—find what people search for, see what ranks, and make something “10x better.” But consensus is precisely what makes you extractable. You were optimizing for a game that no longer exists.

From Volume to Scarcity

So what do you optimize for instead?

Idea scarcity.

To survive the Great Flattening, you must abandon the crowded center of the bell curve. You must run toward the edges—not because the edges are more comfortable, but because they’re the only place where attribution still matters.

You must stop trying to build a taller skyscraper in the city center and instead build a fortress on an island that isn’t on the map yet. You must stop chasing the questions everyone is asking and start defining the questions no one knows to ask.

This requires a fundamentally different discipline:

Observation over optimization. Instead of analyzing keyword difficulty scores, you must go into the field. Talk to customers. Watch how they work. Notice the friction points they’ve learned to accept as “just how things are.” The Tuesday Lag wasn’t found in an SEO tool—it was found in a warehouse.

From the days of the first web browser, the internet economy has been held together by a single, unspoken contract.

It was a simple deal struck between the creators of information and the organizers of information. On one side, you had businesses, publishers, and writers. We agreed to create content—articles, guides, reviews, white papers—and format it in a way that machines could easily read. In exchange, the organizers (primarily Google) agreed to send us traffic.

I give you the answer; you give me the audience.

This transaction was the bedrock of the modern web. It built media empires, launched startups, and created the trillion-dollar industry of Search Engine Optimization (SEO). We built our entire digital lives around the “blue link.” We optimized our headlines, structured our metadata, and obsessively tracked our rankings, all for the prize of being the first click.

But today, that contract is broken.

The Death of the Blue Link

The agent of this destruction is the Large Language Model (LLM). Whether it’s ChatGPT, Claude, or Google’s own AI Overviews, the interface of the internet is changing.

In the old world, if a user wanted to know “how to reduce customer churn,” they searched for it. Google presented ten options. The user scanned the headlines, chose the most promising one, and clicked. That click was the currency of the web. It transferred value from the search engine to the creator. It allowed the creator to capture an email, sell a product, or show an ad.

In the new world, the user asks the same question to an AI. But the AI doesn’t give them a list of links to explore. It reads the top ten articles for them, synthesizes the commonalities, removes the fluff, and writes a perfect, custom summary.

“To reduce churn, you should focus on onboarding, customer success, and regular feedback loops…”

The answer is accurate. It is helpful. It is immediate. And it is catastrophic for the creator.

Because the user is satisfied, they do not click. They do not visit your site. They do not see your brand. They do not subscribe to your newsletter. The AI has extracted the value of your content (the information) without transferring the reward (the traffic).

This is the “Zero-Click” future. And for any business that relies on organic search for growth, it is an existential threat.

The Commodity Trap

The natural reaction to this threat is to try harder. Marketing teams everywhere are doubling down on their old strategies. They are writing longer articles, targeting more specific keywords, and churning out more volume.

But in the age of synthesis, “more” is a trap.

LLMs are prediction engines. They are trained on the vast corpus of the public internet. Their goal is to predict the next likely word in a sentence. This means they are statistically incentivized to produce the average answer. They crave consensus.

If you write a standard guide on “Email Marketing Best Practices,” you are competing with 10,000 other guides that say roughly the same thing. To an AI, your content is just more training data—another drop in the bucket of consensus. The AI will happily ingest your article, mix it with your competitors’, and spit out a generic summary that belongs to no one.

By trying to rank for high-volume keywords with standard advice, you are essentially volunteering to be a ghostwriter for a robot. You are providing the raw materials for your own obsolescence.

Competing in the World of Ideas

So, how do you compete? How do you survive in a world where the “distribution” mechanism (search) is being replaced by an “attribution” mechanism (LLMs)?

In the old economy, the winner was the one with the best distribution. If you ranked #1 on Google, it didn’t matter if your ideas were generic; you got the traffic.

In the new economy, distribution is handled by the AI. The AI decides what to show. Therefore, the winner is the one who gets attributed. The winner is the one who forces the AI to say, “According to [Your Name]…”

You cannot achieve this by being better at the basics. You cannot achieve this by writing “10x content” that is just a longer version of what everyone else is saying.

You can only achieve this by breaking the pattern.

This means creating ideas that are so distinct, so specific, and so named that the AI cannot summarize them away. It means moving from a world of keywords to a world of concepts. From optimization to observation. From traffic to truth.

It means recognizing that in the age of synthesis, the world is dividing into two classes of companies:

The Source: The entities that generate new data, new insights, and new language.

The Echo: The entities that recycle, aggregate, and optimize existing information.

The AI is the ultimate Echo. It can summarize your content, translate it into twelve languages, and reformat it for any audience. But it cannot invent. It cannot go into a warehouse and notice that shipping slows down on Tuesdays. It cannot feel the frustration of a developer managing 1,000 containers. It cannot observe an anomaly in your data and give it a name.

Only you can do that.

Your humanity—your ability to notice what others miss and give those observations language—is your edge. The AI needs you. It needs your vocabulary to make sense of the chaos. And if you provide it, if you have the unique ideas and the unique data to back those ideas, and you package them in unique terminology, then you are not just surviving the AI shift.

You are fueling it. You are the teacher. The AI is the student.

And the teacher always gets the attribution.

This is what it means to build a Linguistic Moat.

There is a quiet conviction I have been carrying in the latter half of my career, and it is this: ideas are the cheapest thing in any room I walk into. They arrive unbidden, they accumulate in decks, they pile up in offsites where leaders nod gravely and then go back to their desks unchanged. What is rare — what is, in fact, almost geological in its slowness to form — is execution. Not execution as a single feat, but as a habit. The capacity to do the right thing not once, in a fit of inspiration, but hundreds of times, on a Tuesday, when no one is watching.

I came up through product and brand. For twenty-odd years I sat in rooms with designers and writers and researchers, arguing about a single word in a tagline, redrawing a navigation flow because the second click felt half a beat too slow, defending a launch story against the hundred small compromises that quietly drain a product of its meaning. That training shaped me in ways I am still discovering. It taught me, above all, that a brand is not a logo and a product is not a feature list. Both are promises. And a promise that the organization cannot operationally keep is not a brand at all; it is a debt, accruing interest in the form of churn and quiet disappointment.

That conviction is what eventually pulled me into operating roles. I noticed that the product and brand leaders who were treated as oracles were not the ones with the most beautiful decks; they were the ones who could trace a creative choice all the way down to a number on a P&L and defend it without flinching. So I made that my posture. I learned to read cohort charts the way I had once read storyboards. I learned to partner with analysts and data scientists the way I had once partnered with illustrators — with respect for a craft that was not mine, and with a refusal to outsource the consequences of their findings. The math was theirs. The accountability became ours, together.

The first thing I did when I stepped into broader leadership was to dismantle a metric. The marketing-qualified lead, the MQL, has been the scaffolding of B2B for two decades, and it deserves a respectful burial. A single human raising a hand on a single form is a poor proxy for what actually happens when a company decides to buy enterprise software. It treats a moment of revealed interest as if it were the relationship itself. The truth is messier and more humbling: a buying group of ten or twelve people, hundreds of anonymous touches, a long preference for peer conversations over sales calls, and a vendor shortlist that is already drawn up in eight out of ten cases before anyone fills out a contact form. If that is the world we live in, then chasing leads is like fishing for a school with a single hook.

So we built our metrics around the account, not the lead. We watched which target accounts were lighting up with research intent, how deeply we had penetrated the buying group inside each one, how quickly accounts moved between journey stages, and which combinations of campaigns they touched on the way. None of this was first-touch or last-touch attribution. Attribution debates, in my experience, are theological. They produce heat and very little light. What we wanted was a picture of the whole orchestra, not a credit line for the violinist — and as someone who grew up writing the score, I found that picture infinitely more honest about what was actually happening on stage.

The second thing I did was to refuse the false economy of cutting brand to fund demand. This is the fight I was born for. Every difficult quarter, the same temptation arrives, dressed as prudence: pull the brand spend, ship more bottom-funnel offers, hit the number. I have watched companies do this for years on end and then wonder, plaintively, why their customer acquisition costs keep climbing and their selling prices keep slipping. Strong brands command premium pricing and lower the cost of every deal that comes after them. That is not a feeling I picked up along the way; it is a finding, repeatedly demonstrated, that the people I came up with happened to know in their bones long before the spreadsheets caught up. So I designed campaigns as full-funnel architectures from the outset, with something to offer the prospect who is not yet ready to register for anything, and I defended brand investment to my CFO with the only language a CFO trusts: efficiency ratios, share of voice, ICP movement from no intent to early intent, the unsexy growth of direct and organic traffic. If you cannot translate your art into arithmetic, you will lose the budget for the art.

The third thing — and this is the part most companies get wrong — was to operationalize all of it. An idea, once again, is nothing. A framework on a wall is nothing. A beautifully art-directed campaign that no one routes, scores, or follows up on within a credible window is nothing. What matters is the management operating rhythm: the weekly meeting, where marketing, sales, customer success, and revenue operations stares at the same live dashboard, in the same color-coded reality, with the same single pipeline goal. Not marketing-sourced versus sales-sourced. One number. One scoreboard. When the early-stage pipeline sagged, marketing stepped in. When closing slowed, sales did. The internal accounting that pits revenue functions against one another is one of the most expensive forms of theater a company can stage, and I have stopped paying for tickets.

Beneath that rhythm, we set behavioral SLAs that were almost monastic in their specificity. Twenty minutes from a qualified-account trigger to the first meaningful sales activity. Three personas engaged within the buying group before the account moved on. Six minutes to respond to an inbound hand-raiser, because inbound intent decays like a half-life. I have seen response times improve by a factor of four within a quarter simply because the dashboard was visible and the expectation was unambiguous. Inspection is not surveillance. Inspection is care.

And then there is the question that hovers over everything now: what does artificial intelligence do to the discipline I have just described? My answer, which I offer without triumphalism, is that it deepens it. AI is a forcing function for the kind of unified data, unified messaging, unified customer journey that I have been advocating for years. The companies that will sound homogenous in the new era are the ones that bolt point solutions onto point solutions and let each one generate copy in its own voice. The companies that will sound like themselves are the ones that consolidate, that train their models on a coherent corpus of who they are, and that insist on a human soul threaded through the automated output. This is, in the end, a brand problem wearing a technical costume.

So I come back, always, to execution. Not as a slogan, not as the bullet point at the end of a strategy deck, but as the actual texture of the work. Execution is the weekly meeting that no one cancels even when the quarter is going well. It is the dashboard that no one is allowed to maintain a private version of. It is the twenty-minute SLA honored at 4:50 on a Friday afternoon. It is the brand campaign that gets defended, in the same breath, with a creative argument and an efficiency ratio. It is the buying group treated as a buying group rather than a list of leads, even when treating it that way takes longer and pays back later. None of this is glamorous. None of it photographs well. But it is what separates the companies whose strategies survive contact with reality from the companies whose strategies are reread, two years later, with a quiet wince. Ideas are easy. Plans are easy. What I have spent a career learning, and what I will spend the rest of it teaching, is the small, repeatable, almost boring discipline of actually doing the thing — every day, on every account, in every meeting, until doing it becomes the only way the organization knows how to operate. That is the discipline beneath the funnel. Everything else is decoration.