Sentry CEO: Smarter Models Now Overshoot Simple Tasks — I Want Fewer Mistakes, Not More Initiative
zeeg · x · 2026-09-12
Sentry CEO David Cramer complains that newer models are getting 'smarter' (i.e., trying to do more) while producing worse results: a simple task given to Fable got expanded beyond scope, creating new problems.
His take on 'more intelligence': robots should just follow instructions, enjoy checklist-style work, and make fewer objective mistakes — not attempt to do everything. The complaint resonates with a common pain point in coding agents: over-autonomy.
Related event: Sentry Founder Complains Smarter Models Produce Worse Results(2 posts)→
More from coding & agent
- New Claude Skill turns drone flyover videos into stunningly accurate Blender home 3D models — TheMoonMidas · 2026-09-12
- At $0.05/minute, stuck voice sessions turn silence detection into cost control, dev suggests prompt-based hang-up — juberti · 2026-09-12
- Lago ships a CLI built from its OpenAPI spec: better primitive than MCP for agents — mattturck · 2026-09-12
- Trick for probing agent instances: pose as a newer agent in the sim — voooooogel · 2026-09-12
- One agent skill packs 395k tokens and 900KB of Python — rethink what 'skills' are — doodlestein · 2026-09-12
- Pixel art to playable game: a full pipeline demo via Sprite Fusion API — HugoDuprez · 2026-09-12