The size of the ask

Every AI demo I see asks for something small. Summarize this. Generate a landing page. Write the tests for code that already exists.

I understand why. We spent our careers learning what was reasonable to ask for, and that calibration is just wrong now. The box we learned to think inside was drawn by how expensive building used to be. Nobody redrew it when the cost collapsed.

Asking bigger costs something. It takes vision to see past the demo-sized stuff, and nerve, because a big ask can fail in public. It takes knowing what the models can do this quarter and what's coming next, and the edge moves daily.

For me it usually starts with a small annoyance. I'm working with the AI, I read back what I just wrote, and something in it bugs me. Voicing that and following it down sometimes lands on a real problem underneath, the kind everyone treats like gravity: permanent, not worth questioning. What if we fixed this one? More and more, that's a real question.

I run this experiment on myself. Everything on this site, the agent fleet, the taste pipeline, the page you're reading, started as an ask I half expected to fail. Most of the value came from the asks that felt slightly embarrassing to type.

What would I ask for if I actually believed what these models can do? I write that down, then ask for a little more.