Seeing things through in the age of AI

AI is fantastic at prototyping. A quick draft of an essay, a mockup of a website, a demo of a video game, concept art or trailer for a movie, or the core argument of a proof - each now takes one prompt instead of a week. 1000x speedup.

AI is useful for helping finish longer projects, but clearly not as good at it. Polishing up a paper draft that would have taken a month still takes two weeks. Finishing a video game that would have taken three years still takes two. Cleaning up every bug, human-checking a proof, covering every use case, handling every small nit and matter of taste, building production-ready infrastructure, all of that is work that, in my experience, only sees a 2x speedup with AI, at best.

AI is a 1000x speedup for drafting and only 2x speedup at finishing. See the problem?

We are seeing an absolute flood of slop in every AI-touched discipline, and nowhere near the promised gains in the stuff that matters. Humans are being completely hijacked by the high of being able to draft every great American novel they ever regretted not writing for a $200 subscription, and bouncing right off the wall of actually polishing any of these novels to a readable standard. The curation infrastructure (referees, editors, moderators) is being taxed beyond all measure, as the tsunami of half-finished demos crashes into its intake. 

It turns out that the best, most impactful, work lives in the tail of the power law in terms of strategy and commitment, and while AI helps with that kind of work, it doesn’t remove the need for deep perseverance.

Written with help from Claude Fable 5.