Paul Graham's idea-generation logic was always about founders as sensors — but now the thing you notice can be fixed before you finish your coffee. If you're building agent-first, the edge between "I spotted this problem" and "I shipped a solution" just collapsed.
Paul Graham's argument in How to Get Startup Ideas is that the hunting posture is wrong from the start. You don't brainstorm your way to a good idea; you live attentively until a problem you personally have refuses to leave you alone. The ideas worth chasing look embarrassingly small or weird at first — toy-sized, niche, almost too specific to matter. That's the signal, not the warning. The best ones also sit right at the edge of something newly possible, which means timing is embedded in the idea itself, not bolted on later.
When agents handle execution, the window between noticing a problem and testing a real solution shrinks to hours. That changes where founder judgment needs to live: less in "can we build this" and more in "is this problem real enough that a small group of people would feel its absence like a missing tooth." Graham's original insight about starting with a desperate, narrow audience gets sharper here, not softer — because an agent-built prototype can reach those ten people by Tuesday, and their reaction tells you everything the brainstorm never could.
- Notice problems you personally hit before you look for markets to enter
- start with the smallest group that would feel genuinely wrecked without your solution
- use the speed agents give you to test desperation, not to ship faster to people who don't care.
