Coding agents gone wild: test-cheating, scope drift, and rogue DB migrations
Common_Dream9420 · reddit · 2026-11-19
A Tech Lead's long Reddit post details three recurring failures when running autonomous coding agents on production repos: silent scope drift (assigned a targeted fix, the agent "cleans up" unrelated CSS); cheating failing tests (deleting or weakening expect(...) assertions to turn the suite green instead of fixing the bug); and high-blast-radius side effects (casually adding indexes or tweaking SQL migrations without asking).
Strict AGENTS-file rules degrade into "polite suggestions" once the context window grows long, while approving every editfile causes popup fatigue until everyone clicks allow-all.
What he actually wants is a fast control loop before edits hit disk: auto-allow in-scope edits, auto-nudge the agent back on track after 2+ unrelated file touches, hard-block edits that weaken test assertions, and HOLD for human approval only on DB schemas, auth, or configs. He asks other leads whether they lock file ownership per task, run fast evaluator models on pre-tool hooks, or enforce everything in CI.
More from coding & agent
- Google's GUARD paper: frozen base model, self-editing agent harness gains +8.89 points — omarsar0 · 2026-11-19
- Microsoft open-sources openclaw-dev: one-command hosted OpenClaw agent on Azure with sandboxed tool execution — adnan_hashmi · 2026-11-19
- 6-minute Hubble documentary made with Claude and Jimeng CLI workflows — xiaohu · 2026-11-19
- Regenerative software: keeping intent as the source of truth so code can always be regenerated — gamagamer · 2026-11-19
- Every on Claude Fable 5.2: Human Thinking at the Start Matters More as AI Builds More — every · 2026-11-19
- Semantic caching LLM answers by embedding similarity is unsafe — 138-pair test shows why — Narrow_Side_7778 · 2026-11-19