Track III · The playbook
Working with
AI agents.
How to read, review, and prompt agents that already write Rust well. Agents already know Rust well; the gap is on your side. When an agent emits 600 lines against your prompt, you do not need to write them from scratch. You need to read them critically, gate them with Cargo, and push back specifically.
The loop
- iPrompt clearly
Name the contract: signature, error type, ownership boundary, tests.
- iiRead the diff
Five minutes, five questions. Diff shape, public API, errors, allocations, tests.
- iiiRun the gates
cargo fmt --check && cargo clippy -- -D warnings && cargo test --workspace.
- ivPush back specifically
Justify each .clone(). Show the error type. Why Arc and not Rc?
- vIterate small
Tiny diffs that pass the gates. Compound the discipline.
Chapters
Six pieces · 45 min total- 01Review lensPer-concept review notes you scan during every diff.The cheat sheet
- 0212 failure modesWhat agents get wrong in Rust, with diffs.12 patterns
- 03Cargo gatesMake agents prove correctness with Cargo.CI + pre-commit
- 04Prompting RustRust-specific prompts that actually work.Templates
- 055-minute PR reviewFive questions, in order, for every Rust diff.5 minutes
- 06Error decoderCommon errors, the lazy fix, and the real fix.rustc decoded