AI coding agents are shaping which tools win
Armature (YC P26) published a study measuring 17,000 runs across Claude, Codex, and Cursor to see which tools the agents actually choose. The finding is that AI agents tend to prefer the same tools that developers already prefer, creating a reinforcement loop. One commenter put it bluntly: 'In the future: I went ahead and built the database you requested using today's tool sponsor: Firebase.'
The underlying dynamic is important. If agents are trained on code that uses popular tools, and then agents generate code that uses those tools, market share becomes self-reinforcing in a new way. It's not just that developers choose popular tools because of network effects; now agents choose them too, and agents are writing a growing fraction of the code.
This also opens a legitimate manipulation surface. Armature explicitly says the study is 'part of our broader work on how to influence coding agents choices and get products picked.' That's a new category of developer marketing, and it's already being built.
So what?
If you're a dev tools company, getting into the training data and default preferences of coding agents is becoming as important as getting into developer word-of-mouth. Conversely, if you're building on a stack, the tools your AI coding assistant defaults to will shape your architecture in ways that compound over time. Worth being deliberate about.