I am seeing increasing unrest around AI pricing, especially after GitHub moved from request-based to token-based billing. The quality of generated code is part of the discussion, but pricing is bringing a different trend into focus.
More people are building private tools. Some are developers generating one-off utilities instead of searching for open source projects. Others are product managers, analysts, finance teams, and operators creating small workflow-specific helpers that would previously have required a software team or a SaaS product.
This raises an interesting question. If everyone can create their own helper using prompts and tokens, how relevant is publishing software for broad reuse, also called SaaS or Software as a Service ?
My view is that AI coding agents are good enough copilots. They can refactor code, implement well-defined tasks, help developers work across languages, identify logic issues, and accelerate debugging. They make developers more productive, but they are not replacing good developers. Domain knowledge, context, and engineering judgment still matter. At the same time, I think developer copilots were just the first phase.
Developers were an ideal proving ground because software is naturally expressed as code. The larger opportunity is helping non-developers create and operate their own workflow tools.
CoWork-style systems are a natural next step. These need working software, but not necessarily long-term maintainability, broad reuse, or support for every edge case. The users are often solving their own problems, for their own workflows.
If AI makes it trivial for individuals to build software for themselves, the remaining opportunity becomes building software that benefits everyone else.
