The capability inventory
A quiet list of what this product can actually do.
Every AI product has capabilities and constraints. Most products only show you the capabilities by accident, when you guess a prompt that happens to work. Everything else becomes a trial-and-error puzzle. A capability inventory is what happens when a team decides their users deserve the full list, written down, in plain words.
The inventory isn't a marketing page. It's a specimen sheet. Each row says what the capability does, how stable it is, and what a working prompt looks like. Users stop guessing. They start picking.
"If you can't name what your product can do, your users will invent it, poorly. The inventory is the cheapest form of alignment you'll ever ship."
One row per capability. Three columns: what, stable, try.
Group by type: reasoning, retrieval, action. Inside each group, the inventory lists what the capability actually does, tagged with a stability badge (stable, beta, unavailable), and a one-line prompt you can click to try. Unavailable rows don't hide. They show exactly why they're dark: waiting on a connector, waiting on IT, waiting on Q3. Absence, well-surfaced, is information.
Tap a row to see what this capability actually does.
Every row is a capability with a named stability tier. Users stop guessing what the model can do.
Users can't use what they don't know exists.
The discovery problem in AI products is brutal. Every model can do more than users think, and every product wraps the model in a shape that hides most of it. The gap between perceived capability and real capability is where retention dies. An inventory closes that gap in five minutes of reading.
Inventories that teach instead of advertise.
- Honest stability tags. Stable, beta, unavailable. Nothing else. Flashy labels like 'premium' or 'new' are for dashboards, not inventories.
- Try-prompts you can click. The inventory teaches by example. The prompt is the shortest path from reading to doing.
- Reasons for absence. When a capability is unavailable, name why. 'Not yet' is a promise, not a wall. Users will wait if they know what they're waiting for.
Inventories that lie about readiness.
The failure mode is marking something stable when it isn't, or writing a one-line prompt that only works for a demo-rigged example. Users will try one of the stable rows, fail, and conclude the inventory is a marketing trick. The inventory then becomes worse than nothing.
What this pattern gets wrong when it gets wrong.
- Confidence theater
- Language or typography that performs certainty beyond what the model actually has.
- Implicit refusal
- Declining a request without saying the word no — a vague answer, a pivot, a change of subject.
Three shipping variants worth copying.
- A sidebar inventory listing every capability with a stability badge
- Each entry has a one-click 'try this' prompt
- Grayed-out entries say exactly why they're unavailable