When AI Disappears, Value Appears

Most writing about invisible AI treats invisibility as a destination, get there and you've won. Using these products myself taught me something different, invisibility isn't a constant state, it's a timing problem, and most products that get called invisible haven't actually solved it.

granola gets the timing right

Granola is the clearest example I've actually lived inside of what invisibility should look like, and it isn't invisible all the time, on purpose. Right as a meeting is about to start, it shows up, a notification, one click to start taking notes and join the meeting at once. Then it disappears. It takes notes diligently in the background for the whole call, and I don't think about it again unless I need something from it mid-meeting, at which point it's easy to pull back up. That's the actual design, visible exactly when I need it to act, invisible exactly when it shouldn't get in the way, on a schedule that matches the meeting itself, not a blanket rule.

wispr flow, the honest counter-example

Wispr Flow gets held up, including by me in an earlier essay, as the model of an invisible AI product, and using it myself, I don't think it actually is one yet. It's the product I most wanted to disappear into my thinking, and it hasn't, not for me. I have real ideas for the product changes that would close that gap, and I'm not ready to lay them out here, they're still taking shape. But I'd rather say that honestly than credit a product with an invisibility it hasn't earned just because the pitch is right.

invisibility is earned, not assumed

The lesson from putting these two side by side is that invisibility isn't a property a product has or doesn't, it's a sequence a product has to get right, when to show up, when to disappear, and when to be reachable again without demanding attention. Granola got that sequence right for a meeting. Wispr Flow hasn't gotten there yet for how I think. Most AI products getting called invisible right now haven't actually earned the word, they've just adopted the pitch.

conclusion

The best AI products won't feel like products, but that's a result, not a starting assumption. It has to be built, moment by moment, not claimed in a positioning line. Granola's proven it's possible. I'm still waiting on the rest, Wispr Flow included.