Since the release of ChatGPT in late 2022, I’ve used AI every day, and for the past year, I’ve been building AI applications full-time.

One thing I’ve noticed is that AI applications are a lot like marketplaces. There’s a clear demand side, a supply side, and an interface connecting the two.

The demand side are the end-users of AI applications. For Cursor, its developers. For Granola, notetakers. For Mixy, DJs. For Particle, newsreaders.

The supply-side for AI applications are LLM tokens. These tokens are good at different things, complete tasks at varying speeds, and come in various price points, like freelancers on Upwork. There’s an ever-increasing fragmentation of tokens, much like books or collectibles before Amazon and eBay aggregated them into a marketplace. Tokens and the foundational model companies behind them want to multi-tenant, meaning listed on different and competing applications, and the end users don’t mind, just like drivers on ride-sharing and delivery marketplaces. 

And just like the most successful marketplaces, the most exciting opportunities for AI applications are in non-consumption, not in competing with human labor. They unlock the liquidity of LLM tokens to enable new experiences and expand the demand for labor that was previously inaccessible. It’s the designer generating code with AI, the nurse hiring an AI notetaker, or a student learning from an AI tutor. 

There’s endless talk and fear about AI eating jobs.