There's a look now. You know it when you see it: the same rounded cards, the same gradient, the same confident-but-generic copy. Teams working with AI design tools routinely iterate two or three extra rounds just to escape it — because the default output has a visible quality ceiling, and users have learned to read the sameness as a warning: this might be a wrapper.
That's the new landscape. Everyone rents the same models, the same component libraries, the same patterns. When everyone can ship a competent product, competence is table stakes — and taste becomes the moat.
Taste, defined without the beret
Taste isn't aesthetic preference, and it isn't decoration. It's judgment under constraints:
- What to leave out
- What not to ship yet
- Which default respects the user's time
- When "impressive" is the wrong choice and "clear" is the right one
A useful test: taste is what's left over when you subtract everything the model generated for you.
Where taste actually shows up now
Not in the hero image. In the corners:
- The error message that protects the user's work instead of apologizing vaguely
- The output that flags its own weak spot: "not sure about the renewal date — worth checking"
- The empty state that teaches instead of decorating
- The feature you didn't build, because it would have made the product harder to trust
- Copy written in the user's vernacular, not the model's
Example
the same card, before and after judgment
Generated: nine options, three icons, a paragraph of enthusiastic copy, every action equally loud.
Edited: two options (the ones users actually choose), renamed in the user's own words, one honest line: "Based on your last 3 invoices."
The second one took twenty minutes of deciding, not designing. That's the moat being dug.
Taste only scales if you write it down
Taste trapped in one person's head has to review every screen — that's a bottleneck, not a moat. Taste written into instruction files ships with every build, including the ones coding agents do while you sleep: the vernacular, the states, what must never happen. (The Founding Designer's Stack shows what those files look like.)
Try this — two exercises that train it
- The stake-your-name pass. Open your product's most important screen. For each element ask: would I stake my name on this? Cut or fix everything that makes you hesitate.
- The credibility audit. Use three AI products this week. Every time something feels credible or cheap, write one sentence naming why. Those sentences are taste being trained — and they're the start of your instruction files.
Go deeper: why moats moved from capability to judgment
Every previous platform shift had a capability moat phase: whoever could build the hard thing won. This shift is different because the hard thing is rentable — the same models and tools are available to your smallest competitor. What can't be rented is the accumulated judgment about your specific users: their words, their anxieties, the moment they decide to trust you. That's why taste compounds while features commoditize.
Capability gets you to the starting line now. Taste — exercised fast, written down, shipped — is how you pull away from the other forty products that started there with you.