craft as ai moat/spring series 4 of 5
ai as substrate. human as director.
ai shortened the gap between a hunch and something you can look at. who decides which hunch was worth having is still a human call.
good design and craft is a feeling more than a skill you can hand to an ai: a hunch built from experience and expertise.
that feeling does most of the invisible work on every product you actually love. it is why linear feels different than a competitor that does the same thing. it is why some landing pages make you want to click and some just sit there. the effect comes from the accumulation of a thousand small decisions, each made by someone with a strong sense of what right felt like.
ai has not rewritten that job. what it has changed is the speed between having a hunch and seeing it alive.
the workflow that works for the best builders we know looks nothing like a textbook methodology. idea, prompt, prototype, then a spread of options across models, then a cut back down to one, then a lot of refining by hand. from the outside it looks chaotic. from the inside it is chaotic, and it still works.
you have an idea. instead of writing a brief, you prompt. you generate three to five divergent versions immediately: actually different takes, not slight variations of the same concept. you compare results across models, because claude's option space is genuinely different from flux's or gemini's. the differences between models are themselves a form of creative input.
then leave the first thing that looks good alone long enough to compare. look at all of them and see which one has the spark: the direction that feels different. pull that into your hands to finish.
finishing by hand is where craft shows up. spacing gets intentional. transitions stop janking. copy sounds like a person wrote it because a person did. dark mode covers the edge case. the empty state gets the treatment it deserves.
ai is the substrate. you are the director. the same relationship a film director has to a cinematographer, a sound designer, a production designer: all of them talented, all producing technically excellent work. the director's job is the taste layer. which take lands, which angle cuts, what the whole thing should feel like.
the math is real. a $10m arr business that used to need 50 to 70 people now needs 15 to 20. the minimum viable team is shrinking fast. a design-led founder, a technical co-founder, someone who can tell the story: that is a $10m team in 2026.
notice what those people actually do. they do not type faster than a 200-person org. they direct better. they have a stronger point of view about what is worth building and how it should feel. ai handles the lifting. humans handle the taste.
the role did not disappear. it evolved into builder, director, tastemaker. the job title is the least interesting part. judgment is the part that still has to be staffed.