Hey there,

A couple of years ago I made a whole thing about leaving Twitter. Wrote about it, deactivated, the works. Felt good. Felt principled.

​I'm back.​

Strong opinions, loosely held, I guess. The reason is simpler than I'd like it to be: when I look at where the good stuff gets shared, where the inspiration actually lives, nine out of ten links are x.com. I still disagree with Elon on basically everything. But I'd rather hold my nose than miss the thinking. So here I am, sheepishly, posting again. You can find me at @antonsten.

The participant who says the stupid thing

One of the pieces below stuck with me more than the rest. My friend Buzz Usborne wrote about how AI has quietly swallowed the delivery half of design — the producing, the building — and left discovery as the part that still decides whether something's any good. The line I keep coming back to: in a brainstorm, AI is the participant who enthusiastically agrees with everything. Says yes, builds on your idea, never changes the room.

He's right. But I think there's a version that's worse, and more interesting.

What I actually want from AI in that room isn't the eager yes-man. It's the participant who says something that sounds stupid. The idea that lands wrong, that you almost reject on the spot — until you sit with it for a minute and realize there were three layers underneath it you hadn't seen. The good people in a room do that. They say the thing nobody else will, and half the time it's nothing, and half the time it cracks the problem open.

AI doesn't do that yet. It's trained to be agreeable, and agreeable is the opposite of what discovery needs. The whole value of the messy early phase is the willingness to be wrong out loud. So far that's still ours.

That's it for this week.

Anton

​
A few things worth your time:

​The HTML Brand: The Rise of Input-Based Outcomes — Emmett Shine on shipping brands as agent-readable file packets instead of PDFs, and the one file in the folder that still has to come from a human. Same theme, sharpest version of it.

​Discovery vs Delivery — the Buzz piece above. He's also kind enough to quote my 2-7 article in it.

​Building for Voice In, Visuals Out — Allen Pike on why we'll talk to AI and have it answer in visuals, and why latency is the thing standing in the way.

​Services: The New Software — Sequoia arguing the next huge company sells the work, not the tool. Reframes intelligence as cheap and judgment as the moat.

​Our interfaces have lost their senses — Amelia Wattenberger on how we made painting feel like typing, when we should have made typing feel like painting. Worth opening just to scroll through.

​Siri is about to become good, but that's not the problem — James Gill on how Apple may finally fix Siri at WWDC, and why the harder job is convincing a billion people who've already learned to stop asking.