If you're delegating work to agents, you're also generating a live feed of friction — and that friction is where the next company lives.
Paul Graham's essay on startup ideas is, at its core, an argument against imagination and for attention. The mistake founders make is sitting down to think up ideas rather than training themselves to notice what's already broken in their daily lives. The best ideas, he argues, emerge from being the kind of person who lives at the frontier of something new, feels a genuine want, and trusts that if they want it, others quietly do too. Small initial markets aren't a weakness — they're a signal of precision.
When agents handle your workflows, something interesting happens: the seams show. Every place an agent halts, asks for clarification, hands back to a human, or produces output you silently fix — that's Graham's essay playing out in real time. The embarrassing workaround your team built last Tuesday is the toy product he's describing. Founders in AI-native studios should treat their own agent stack as a problem-discovery instrument, not just an efficiency tool. The next startup idea is probably already living in your correction logs.
- Watch what your agents can't finish without human rescue
- trust the tiny audience that desperately needs the fix before you chase the large one that might.
