LLMs make it very easy to develop sophisticated solutions for increasingly narrow problems. However, it’s not as easy as just building them. Working with these agents takes some skill, and the teams with products in market are already using these AI agents to bolt on features faster than you can blink. More accurately, LLMs haven’t democratized building in the sense of empowering individual developers, but rather have made niche markets economically viable for established companies by dropping development costs.

In the past, there were serious constraints about market size and the “vitamin vs painkiller” perspective (venture capitalists prefer painkillers over vitamins, apparently). The conventional wisdom is that you should only build painkillers: it’s not worth it otherwise. The market for vitamins is too small and a not enough people care about vitamins enough to adopt your product. That’s not as true anymore.

First, the internet created viable niche markets, but development costs kept established companies (or anyone really) from serving them. Take Facebook Groups, think of whatever random niche interests, there’s a FB Group for it, and it has several thousand members. Those people have preferences that have been niche, and they have been ignored, at least commercially. The niches have been big but LLMs unlock them in an interesting way.

LLMs removed this cost barrier. The cost of producing for these niche audiences has dropped. You can build (almost) anything. While it’s true that distribution, customer acquisition, and maintenance costs remain substantial for niche products, these products tend to have a head start leveraging these same communities. The low cost of development is also seductive to individual developers, driving sky high valuations for “prompt-to-app” services like Lovable and Bolt.

The side effect of all of this is a consumer surplus from increased variety. Unsatisfied with the choices from when I last looked a few years ago, I was halfway to building a wine tracker app before I discovered a new one that had all the features I needed and more. My fantasy football app integrated AI features faster than I could blink. It’s clear that the pace of development is accelerating, especially for the services that users already have.

Finally, what’s most interesting is that these opportunities are not being won by individual developers or upstarts, but empowering existing companies with infrastructure, distribution advantages who can move quickly to bolt on features.

To get started with Claude Code, try this deeplearning.ai course