A screen for category-defining companies

The Filter.

A weapon, not an essay. Run any idea through the six gates in five minutes. Fail two, it's a toy. Pass all six, it's rare — and rare is what wins.

Don't disrupt an incumbent. Create a category that didn't have a name yet — and be the first to make it magic.

Enter through the individual. The engine doing the work — today AI, tomorrow whatever's next — is a commodity. Everyone has it. Your edge is never the engine. It's always what you build on top of it.

This is Phase 1 — how it gets created and takes off.

The Six Gates

06 gates
01

New category, first to make it magic

Being first is not a moat. Friendster and MySpace were first and got erased. What compounds is being first to love-at-first-use — first to PMF, not first to ship.

In plain terms

Don't fight for share in a market that already exists — mint one that doesn't. But 'first' is worthless if the first version is forgettable. You're not racing to a launch date, you're racing to the magic moment — the version people can't put down. Get to product-market-fit before the rest of the market even realizes the category is there, and the category ends up with your name on it.

The test

Are you opening a space that had no name, and can you be the first to make it magic — not just first to arrive?

PassesElevenLabs. First AI voice good enough to give you chills.
DiesAny first mover shipping mediocre, holding the door for whoever makes it magic.
02

A 10x wedge on one thing

Enter through the individual, but only with something 10x better at a single, concrete thing. And if the engine is a commodity, your 10x cannot come from the engine.

In plain terms

Pick one job and be an order of magnitude better at it — not 20% nicer across ten features. Ten-percent-better doesn't get shared or remembered. 10x does. And since every competitor rents the same models you do, your 10x has to come from somewhere the model can't hand out: a workflow you compressed, a latency you killed, an insight into the job nobody else framed. If the only answer to 'why are you 10x?' is 'the model is good,' you have a wrapper, not a wedge.

The test

What is your 10x in one sentence — and where does it come from, if not 'the model is good'?

PassesLovable. Idea to shipped app in a minute. The 10x is speed and flow, not the model.
DiesThe wrapper whose only pitch is 'we use GPT.' A demo, not a wedge.
03

The output is its own ad

If the output shares itself, distribution lives inside the product and CAC collapses. But it must land on the next person who'd also use it, and arrive warm — recommended and already primed for them.

In plain terms

The cheapest growth is a product that markets itself, so distribution should live inside the product, not inside a sales team. Two flavors count. Either the artifact is literally shareable — a voice, a video, an app that carries your product with it — or growth rides bottoms-up word-of-mouth and multiplayer, a tool teammates pull each other into. Cursor's output (code) isn't an ad, yet it spread desk to desk. Both collapse CAC; a category with neither is buying every user, forever.

The test

Does distribution live inside the product — a shareable artifact, or bottoms-up word-of-mouth and multiplayer — landing warm on the next likely user?

PassesShare an ElevenLabs voice with someone who'll also make voices. The artifact is the funnel.
DiesA generated résumé, shared with a recruiter — not another job seeker. Doesn't propagate.
04

Craft, it has to feel great

When everyone runs the same engine, taste wins, not tech. Design, speed, polish, the thousand small decisions. Can't be bought with capital or compute — it's judgment.

In plain terms

The moment everyone can call the same model, the model stops being the differentiator and taste takes over. Two products with identical features are not equal — the one that feels effortless wins, keeps users, and earns the word-of-mouth. That's speed, feel, the details, the restraint to leave things out. It's the one moat that scales with judgment instead of budget, which is exactly why a well-funded competitor can't just buy their way past it.

The test

Does it feel great to use, or does it merely work?

PassesMidjourney on aesthetics, Linear on speed and feel — same class of tech as rivals.
DiesIdentical features, but it feels slow, ugly and plastic.
05

Frequency, do they come back tomorrow?

Half of viral AI toys die here: big spike, zero habit. A product used once and abandoned is not a company. Category-definers are habits.

In plain terms

A spike is not a business. Virality gets you the install; a reason to return gets you the company. Ask where the product sits in someone's week — is it a habit with a natural trigger, or a novelty they try once and forget? Frequency is also what powers everything else: only repeat use lets memory accumulate, the data flywheel spin, and retention compound. No habit, no compounding, no category.

The test

Is there a natural reason to come back every day or every week?

PassesCursor. You return every time you code — which is daily.
DiesThe novelty generator you try once and never reopen.
06

The unfair "around"

The mandatory question: what do you own that is not the commodity? Craft, a dataset only you have, a workflow only you understand, a loop only you spin. If the answer is 'nothing,' you have two guns to your own head.

In plain terms

The model is a commodity everyone rents, so your durable edge is whatever you build around it that a competitor with the same API can't copy. Proprietary data that accrues with usage, a multiplayer graph, a workflow the org can't rip out, a brand people trust. Name it out loud. If you can't point to something you own that isn't the model, you're renting your entire moat from OpenAI — and they can evict you the Tuesday they ship your feature.

The test

What do you own that nobody can copy even with the same engine?

PassesFreepik. On top of the wrapper: a library, a brand, teams collaborating inside.
DiesThe pure wrapper. The platform ships your feature on a Tuesday and you evaporate.

Scorecard

0/6

Verdict

Run an idea