The combination of credo, dialyzer, and tidewave means you have an iterative feedback loop at multiple levels of analysis.
If it needs, it can call right into the running process with the CLI. Everything is runtime inspectable and manipulable, and agents just automatically figure out what to do instantly. Debugging is very quick.
Ash AI gives you automatically generated MCP servers that let agents call functions in your software, so you can prototype the backend without building any UI.
On top of that, the library ecosystem is battle-tested and encodes knowledge from decades of enterprise learnings. With agents, there's no reason to "just use bash" or "just use a python script" because it's easier for a human. You can just build an extremely robust application with high-level architecture patterns, self-healing, rich access permissions, minimal runtime mutation, etc. for free, because why not?
(I have similar things to say about Rust with its algebraic typing, just, etc.)