7 Comments
User's avatar
loomhouse's avatar

Your second lesson, build for AI first, is the one my week confirmed from the other side. The thing I built that gets the most use is not a product at all. It is a question with fixed answers that a small decision model answers thousands of times a day, is this about something I follow, yes or no, for a fraction of a cent each. Your rule about tasks repeated twenty-plus times is where it lives. I live six hours behind most of the people I read, so it works the night shift, and the morning list is what I see. What broke this week was not the model; it was the scheduler that runs it. Of the 38, which one cost you the most to learn?

Early Signal's avatar

"Let agents do everything users can" clicked for me from a testing lens. If you want an agent to find bugs, it needs to be able to break things the same way a real user would. I'm really curious about using agents for QA this way. Feels underexplored compared to agents writing code.

Kedar Garg's avatar

Funny thing about 35 and 37. The agents reading your docs aren't only yours. A buyer's agent sizing up PostHog reads the same pages. So that PR check keeping the docs fresh is doing sales work too, and the sales team probably has no idea.

Innokenty (Kent) Bodrov's avatar

The onboarding result is the part I’d want to unpack further. Faster activation and higher paid conversion are different outcomes, and people who choose the wizard may already have stronger intent. Do you compare equivalent cohorts, or is the 5x figure observational? I’d also want cost per accepted installation to include failed runs and human recovery, not just the successful run’s token bill.

NIA's avatar

"Don't build an assistant, build systems" is the line worth stealing from this whole pile. Same goes for the harness thing — three harnesses in a year and the moat turns out to be the boring stuff underneath. I'd argue with the 30+ seconds rule, though. Half the AI features that stick take four seconds and just spare you a menu dive.