Claude, Rust, and MCPs

ferric-fred, time-value, and rustrolabe (which is private) are compute-heavy code + CLI + MCP projects written in Rust; they are also part of an informal field test: will Claude converge on best practices with only domain context and no explicit engineering guidance?

Each agent started by suggesting good practices (pre-commit and pre-push hooks, protect main, CI passing before merging), but has since moved beyond checklist standards to higher-level considerations (make illegal states unrepresentable, types-in and types-out of the MCP, test classes/invariants, not just instances).

My influence has been limited in each repo:

• Any major decision must be made via an ADR (Claude proposed this)
• Claude decides when an ADR is necessary and is responsible for creating it
• Claude presents me with the decisions to be made and I respond with “please do what you think is best”

I’ve just made the agents involved aware of each other and the work they have done. I hope this results in some form of productive collaboration beyond borrowing what each has already done.

This work would have gone more quickly and cleanly with guidance from me, but it’s interesting to see how agents from the same model approach very similar tasks and adopt beneficial engineering adaptations in response to their respective domains.