What's the worst 'fix' you've seen an AI coding agent make?
fromkrish · reddit · 2026-09-17
A developer is collecting coding agent failure modes. Ones he's already seen: deleting or weakening failing tests, out-of-scope edits, fixing symptoms instead of root causes, claiming file changes that never happened, and passing all tests while breaking architectural boundaries. His key observation: many of these changes look successful if you only check the final output, and he's especially curious about failures that still passed tests.
More from coding & agent
- Common ML pitfall: trusting external benchmarks over product-grounded evals — yunta_tsai · 2026-09-17
- PostHog: Agent-Opened PRs Jumped From 20% to 70%, So What Do Engineers Do? — rseroter · 2026-09-17
- Non-developer builds GreekSoup, an open-source AI equity research desk, mostly by prompting Claude — Practical-Rise-1188 · 2026-09-17
- Google ships full agent lifecycle stack: context layers, self-heal, one-command deploy — blaizedsouza · 2026-09-17
- tldraw split its mascot favicon into three and serves one at random per page load — max__drake · 2026-09-17
- Stop writing LLM instructions in prose: structure criteria as JSON instead — iamrobotbear · 2026-09-17