This deck started out of curiosity. How did the AI app builder space begin? Which part of the stack is actually capturing the money: models, infra, or builders? Why are the feature sets of these products converging so quickly? And what explains why some builders go viral faster than others?
After digging in, a few themes stood out:
Models and infra capture the rents
If you zoom out, the economics of the stack are very clear. Models like OpenAI and Anthropic have predictable usage-based revenue. Infra players like Supabase and Vercel are AI-agnostic and capture value no matter which builder wins.
Builders, on the other hand, live with thinner margins. As Nicholas Charriere, founder of Mocha, put it: “Margins on all of the code gen products are either neutral or negative. They’re absolutely abysmal.” (link) That pressure is already reshaping strategy.
From wedges to convergence
Each builder began with a narrow wedge: Lovable excelled at polished frontends, Devin specialized in enterprise workflows, Tempo focused on Figma-to-code. But users soon demanded full stack solutions. Lovable shows the shift as it added Supabase for data, Clerk for auth, Stripe for payments, and GitHub for code sync to move beyond its frontend wedge into a full stack platform.
The network effect that drives virality
Expanding the product was only half the story. What really made Lovable break out was distribution through community. Templates lowered the barrier to entry, remixes let users adapt each other’s work, and that cycle pulled in more users. The template-to-remix culture became a self-reinforcing loop of growth, and that is what pushed Lovable to go viral.