Phoenix Creator Argues Elixir Is AIs Best Language (thenewstack.io)

At ElixirConf US 2025, Phoenix creator Chris McCord argued that Elixir — powered by the BEAM VM and the Phoenix framework — is uniquely well suited to be the "frontend" language of an agentic AI web. He demonstrated a Phoenix-based AI agent that scaffolds a Slack clone and introduced an AGENTS.md file baked into new projects. AGENTS.md encodes architectural context and explicit coding heuristics for LLMs (e.g., avoid index-based list access), smoothing over common model mistakes so agents produce correct, idiomatic Elixir. McCord also showed a Phoenix agent that self-summarizes and “self-collapses” its context window so it can sustain long-running tasks without losing track. The talk matters because it reframes language choice for agentic workflows: Elixir’s concurrency model, in-memory system index on a GenServer, filesystem monitoring that streams file changes to the LM, and cohesive tooling (Mix, Phoenix.New, Tideway Web) solve many practical agent problems (caching, GC, multi-core orchestration) that other ecosystems are still wrestling with. While LLMs tend to prefer JavaScript because of training data, McCord argues that Elixir’s unified stack and explicit agent guidance can make it the better platform for reliable, long-lived coding agents — a timely claim as agent capability and task-horizon appear to be growing rapidly.