'Fix the bug' is a bad agent task: separate expected behavior, evidence, and acceptance
Reasonable_Ant_8356 · reddit · 2026-11-22
Many agent failures start before the agent runs. The author argues a task like "fix the pagination bug" is underspecified — duplicate records, empty-result crashes, and wrong page counts after filtering are different problems. The fix: structure task descriptions into expected behavior, evidence (tests covering edge cases), and acceptance (AI review plus human sign-off), so someone other than the executing agent can verify the result. Mirrors Sodu Work's request workflow.
More from coding & agent
- DeepSeek Harness v0.2.3-alpha.1 ships 14 changes: plugin market, PR review, multi-agent UI — teortaxesTex · 2026-11-22
- 2-day Gemini stress test: $4.99 Google AI Pro nearly maxed, orchestrator UI revealed — lipeng0820 · 2026-11-22
- Graffitty: open-source wide terminal emulator with remote access, 3D backgrounds, graph orchestration — kiyoonkim · 2026-11-22
- Redditor argues AI agents need decision expiration dates, not just bigger context windows — Elvantime · 2026-11-22
- ianand's 20-min AI Engineer World's Fair talk covers a paper and how it can go wrong — js_craft_hq · 2026-11-22
- Agent pipeline silently skipped a midnight run due to a UTC vs local time mismatch — fantomkoenig · 2026-11-22