Coding agents write defensive code with overly broad exception catches, and review agents still LGTM it
srchvrs · x · 2026-09-28
prajdabre vents about coding agents' poor test quality and over-defensive code: everything gets wrapped in try-except clauses catching exceptions so broadly that failures become silent — and a review agent will still read it and say "magnificent, LGTM."
Quoting that, srchvrs adds nuance: he finds agent-written Python quite good overall, but intent drift is real — they implement what's not needed and skip what's necessary. As he puts it, you can't extract 1000 bits of code from 1 bit of input without heavy assumptions.
More from coding & agent
- Cursor Launches Self-Hosted Machines: Turn a Mac Mini Into a Remote Agent Worker in One Command — mattyp · 2026-09-28
- doodlestein says he now ships 2,000+ commits per day with agents — tokenbender · 2026-09-28
- doodlestein's workflow: beads as north star plus a 'reality check' skill to verify code matches plans — doodlestein · 2026-09-28
- Learning Fusion 3D, Then Letting Claude Design Mechanisms for 3D Printing — alfcnz · 2026-09-28
- Open-source skillranker ranks Claude Code agent skills by live session context — doodlestein · 2026-09-28
- RRSI: Regularizing Recursive Self-Improvement Boosts OOD Agent Benchmarks Up to 4.7 Points — burny_tech · 2026-09-28