# Fly.io pitches Phoenix.new as a remote AI runtime for Elixir
Fly.io is introducing Phoenix.new as a remote AI runtime for Phoenix, the Elixir web framework, after what its creator describes as a skunkworks effort that started as a weekend project. Chris McCord says he wanted LLM agents to work with Elixir as well as they do with Python and JavaScript, and that goal pushed the project far beyond the original experiment.
The basic pitch is straightforward: give Elixir developers a full environment in which AI agents can build and run applications without treating the language as an edge case. McCord says he began in December trying to understand how difficult it would be to build a coding agent in Elixir. A few weeks later, he had a system that could produce working Phoenix applications and drive a full in-browser IDE. That is the point at which the side project became something more ambitious.
Phoenix.new is being framed not just as a demo but as a batteries-included runtime for building with AI. The wording matters because it suggests a product that wraps together infrastructure, development tooling and agent interaction rather than leaving each piece to be assembled separately. For developers in the Elixir ecosystem, that could reduce one of the major frictions in AI-assisted software work: getting tools meant for Python or JavaScript to fit a different stack.
The announcement also says something about Fly.io's positioning. The company is known for infrastructure and developer tooling, and Phoenix.new lets it show off in a part of the stack where runtime, application framework and agent execution all intersect. By presenting the project as a formal introduction rather than a tentative prototype, the company is signaling that it sees real product potential in the idea.
McCord's comments make clear that the project was not conceived as a research stunt. He says the work had been teased publicly on Twitter as it picked up steam, and that the team has now reached a point where it is serious enough to introduce to the world. That shift from private iteration to public launch is often where hobby projects either stop or become platforms.
The use of a full in-browser IDE is especially notable. It suggests a system designed for immediate interaction, not just background code generation. If agents can reliably produce Phoenix apps and developers can inspect and adjust them in browser, the runtime becomes a bridge between AI assistance and conventional software work rather than a separate toy environment.
For the Elixir community, the launch is also a vote of confidence. It says the language is still worth investing in as AI changes the software stack. Phoenix.new is not claiming that LLM agents require Elixir, but it is arguing that Elixir should not be left behind as the tooling around agents evolves.
The project also fits neatly into a broader shift in software tooling, where AI features are increasingly expected to work across the full stack instead of only inside the most common languages. By making Elixir part of that story, Phoenix.new is trying to show that runtime choice does not have to dictate whether agent-based development is practical. The pitch is not that Elixir replaces Python or JavaScript, but that it deserves equal treatment when the tooling gets serious.
