Ted Hayes built Flash microsites at a small Manhattan agency in 2006, and he uses that era to make a point about AI. Flash was genuinely bad — inaccessible, unskippable intros, broken back buttons — and it also enabled a decade of people making strange, new things because the tool made experimentation cheap. Roughly 98% of web-connected PCs had the player by 2006, and the Flashpoint archive still catalogues 157,509 preserved works. Most of it was junk. You do not get new good ideas without sifting through piles of bad ones.
Today’s tooling makes production cheap again, but not experimentation:
- Around 35% of newly published websites were AI-generated by mid-2025, per an April 2026 study
- The output is competent, legible, accessible, and nearly identical — because LLMs are trained on the settled internet
- The difference from Flash is the humans: nobody making the important creative decisions is trying anything
- The competitive pressure is real, not imagined — Y Combinator application volume moved from 20,000 to 27,000 per batch, and YC went from two batches a year to four
His argument is that a single novel idea is not a moat, because it can be copied. The moat is the capacity to keep inventing: spotting overlooked problems, reframing familiar things, making unusual choices coherent, and continuing to evolve after competitors catch up. That capacity is a habit, not a talent, and the habit has one requirement — letting an idea live long enough to find out whether there is anything in it.
The practical instruction is small and specific: next time you are working a product idea with a model, ask it to surface the assumptions you are making, pick one, and work out what it would mean for that assumption to be wrong. Expect some of the response to be nonsense; deciding what is likely true is the exercise. Use the speed for more experiments, not more replicas. As he puts it, a competitor cannot acquire all of that by prompting Claude.