Building an Autonomous Coding Agent in Rust: Architecture, Decisions, and What I Learned

I have been building Akmon for several months — a terminal AI coding agent that ships as a single Rust binary. No separate runtime, no package manager, no installer. Copy the file and it works. This is not a "here is my project" post. It is an honest account of the decisions I made, the tradeoffs involved, and the things that surprised me. If you are building in the agent space or are curious how autonomous tool-calling loops actually behave in practice, I hope it is useful. Why Rust The choice was pragmatic, not ideological. I needed one artifact that behaves identically on a developer's MacBook, a Linux server accessed over SSH, a Docker container in CI, and an air-gapped environment with no internet access. Rust's static linking story and lack of a managed runtime match that deployment model directly. The release binary uses LTO, size-optimized settings, and stripping. The result is 3.4 MB that runs anywhere you can run a normal executable. The second reason is structural. An agent