Architectural drift at scale
Your team has shipped fast with agents for two quarters and now 70% of the new code is machine-written. The codebase still passes tests, but you've noticed drift: three different HTTP clients, two date-handling styles, validation logic copy-pasted into every handler instead of shared, and modules that each look fine but don't compose. Each PR was individually reasonable. As the staff engineer, how do you keep architectural coherence when most code is machine-written and no single agent run sees the whole system?
find_convention_violations(usage_lines: list[str], canonical_lines: list[str]) → list[str][["billing|http|axios","billing|date|dayjs","search|http|fetchWrapper","reports|http|axios"],["http=axios","date=dayjs"]]out["search|http|fetchWrapper"][["a|validation|inline","b|validation|shared_validator","a|logging|winston"],["validation=shared_validator"]]out["a|validation|inline"]Treat the AI’s output as a draft to verify, not an answer to trust. Name the specific flaw and the input that triggers it, say how you’d catch it (tests, edge cases, reading critically), and how you’d re-prompt or decompose to get it right.
Vibe & agentic: describe the solution in plain language (or narrate it) and the coach grades your approach.