Most early-stage AI tools should join an existing category, not create one. Creating a category means you own the language and set the terms, and the handful of companies that pull it off capture an outsized share of the value. It also means paying to teach the market that a problem worth naming even exists, and that education bill is enormous. Joining a known category costs you the ability to define the game, but it buys instant comprehension: a buyer reads your homepage and knows what shelf you sit on in three seconds. Comprehension is the thing you cannot afford to lose at seed stage. So the default answer is to join, position sharply inside the category, and earn the right to expand it later. The exception is narrow, and this post is about how to tell whether you are the exception.
What does “category design” actually mean
Category design is the work of deciding what your buyer thinks your product is before they compare features. It answers one question in the buyer’s head: “what kind of thing is this.” If the answer is a familiar one like “LLM observability” or “vector database,” you have joined a category. If the answer is a phrase nobody has heard, you are trying to create one, and that phrase now has to do a second job of explaining an entire problem class from scratch.
People conflate category design with a clever tagline. It is not that. The market category is one of the five components in April Dunford’s positioning framework, and it sits last on purpose, because it depends on the others. Your category is the context that makes your differentiator obvious to the right buyer. Get the differentiator and the competitive alternative right first, and the category question mostly answers itself. I walked through that ordering in how to position a technical product without dumbing it down, and it matters here because a made-up category with no differentiator underneath it is just a made-up word.
Why does creating a category look so tempting
Because the winners look incredible in hindsight. The classic argument comes from an HBR analysis of the fastest-growing companies: the small share that created their categories generated 53 percent of incremental revenue growth and 74 percent of incremental market-cap growth among the group (How Unicorns Grow, HBR). Own the category and you own the economics. Datadog trained a generation of buyers to say “observability.” Snowflake made “cloud data warehouse” a line item. When it works, you are the default, the analyst reports get named after you, and competitors spend years positioning against your language.
Here is the catch that survivorship bias hides. That 74 percent figure describes the survivors, the companies that already won. It does not count the far larger pile of startups that coined a category, spent their runway teaching the market, and ran out before the market caught up. Category creation is a high-variance bet. The upside is real and the failure rate is brutal, and at seed or Series A you usually cannot fund the multi-year education campaign the bet requires.
What is the real cost of coining a new category
The cost is comprehension, paid daily. When your homepage names a category the buyer has never seen, every visitor has to do unpaid cognitive work to figure out what you are. Most will not bother. This is fatal now that buyers self-serve almost the whole way: Gartner found 61 percent of B2B buyers prefer a rep-free buying experience and run their own research before ever talking to you (Gartner, via Demand Gen Report). Your invented category has to teach itself while you are not in the room. If it cannot, the buyer bounces to a competitor whose category they already understand.
The costs stack up in specific ways:
- Search demand does not exist. Nobody types your new category into Google or asks an answer engine about it, so your SEO and AI-search visibility have nothing to attach to. You are creating demand, not capturing it, and that is a different and slower motion.
- Sales cycles get longer. Every deal starts with “so what is this,” which means educating each buyer individually, which does not scale until the market has done the learning for you.
- You fund the whole classroom. If a competitor’s product fits your new category too, your education spend teaches buyers to want the category and then they comparison-shop. You paid to grow a market your rival also harvests.
For an AI tool selling to engineers, there is a sharper version of this problem. Your buyer runs their own evals before they trust a word you say, which I got into in marketing to AI engineers. A skeptical technical buyer handed an unfamiliar category label does not lean in with curiosity. They read it as marketing that is avoiding a straight answer, and straight answers are the currency with this audience.
When is creating a category actually the right call
When you genuinely cannot fit an existing category without lying, and you have the runway and distribution to teach the market. Both conditions have to hold. A real new category usually shows up as a buyer who has no name for their problem yet, is solving it with a duct-taped mess of tools, and lights up when you finally describe the pain precisely. If your best customers keep saying “I didn’t even know this was a thing you could fix,” that is a signal worth taking seriously.
The runway test is just as important. Category creation is a marketing-heavy, multi-year commitment, so it fits a company with real funding, a founder who can command a stage and evangelize, and ideally a design partner base already living the problem. If you are a three-person team with eighteen months of runway, you do not have the budget to run a classroom for the whole market. Spend that runway getting understood, not getting studied.
The play most AI startups should run: wedge into a known category, then expand
Enter through a category buyers already understand, win a sharp beachhead inside it, then stretch the definition once you have customers and credibility. This is the move that gets you comprehension now and optionality later. You show up on a shelf the buyer already shops, so the first conversation is about whether you are better, not about what you are. Then, as you accumulate proof and usage, you widen the frame toward the bigger idea you actually care about.
Concretely, that looks like leading with the known category in your headline and reserving the expansive vision for the layers underneath. A tool that wants to define “agent reliability” someday can launch as “LLM observability, built for agents,” which borrows all the comprehension of observability and plants a flag on the part you own. You get found, you get understood, and you set up the expansion without betting the company on a word nobody knows yet. When you do launch, that clarity is what makes the announcement land, which ties into the devtools launch playbook.
The risk to manage is competing purely on features once you are inside a crowded category. The answer is not to flee to a made-up category, it is to niche down hard. “Observability for teams running LLM agents in production” is a known category with a specific wedge, and that specificity is what keeps you from being one more logo in a comparison grid.
How to decide, by stage and budget
Match the strategy to what you can actually fund and how much comprehension you can afford to lose.
| Your situation | Join a category | Wedge then expand | Create a category |
|---|---|---|---|
| Stage | Pre-seed to Series A | Seed to Series B | Series B+ with a category thesis |
| Budget for market education | Little to none | Moderate | Large, multi-year |
| Buyer awareness of the problem | High, named already | High pain, partial language | No name for it yet |
| Primary risk | Feature-by-feature comparison | Expanding too early | Running out of runway teaching |
| What you optimize for | Instant comprehension | Comprehension now, ownership later | Owning the language |
Read the table top to bottom for your stage, then be honest about the budget row, because it overrides everything above it. The most common expensive mistake I see is a seed-stage team with a genuinely novel product deciding they are a category creator, writing a homepage in a private language, and then wondering why trial signups are flat. The product was fine. The category bet was unfunded.
If you are hiring your first devtools marketer, this is one of the first calls they should pressure-test with you, because it shapes the messaging, the content roadmap, and the search strategy all at once. Get it wrong and every downstream asset inherits the confusion.
The short version
Join a known category, position sharply inside it, and expand the definition once you have earned it. Creating a category is a real strategy with real winners, and it is the wrong first move for almost every early-stage AI tool, because the education cost lands before the revenue does. Pick the category your buyer already knows, plant your flag on the specific wedge you own, and let comprehension do the compounding.
If you want help figuring out whether your AI tool should join a category or coin one, and turning that call into positioning your buyers actually repeat, that is the work I do. You can get in touch here.
