Dev finds non-thinking LLMs 'useless' for agentic work in instruction-following tests
madcow_bg · reddit · 2026-11-10
A Reddit developer building semi-autonomous agents for private data sorting reports that LLMs without thinking mode — including Qwen3.6, Qwen3-coder and gpt-oss — consistently fail to follow detailed AGENTS.md instructions, with any slightly vague directive ignored entirely. He concludes non-thinking LLMs are useless for agentic work and asks whether a systematic study on the importance of thinking exists.
More from coding & agent
- Math explainer video on extreme value theorem built in 40 minutes with Codex — cloneofsimo · 2026-11-10
- Use mind-shaped priors when working with LLMs, argues researcher — Sauers_ · 2026-11-10
- Busy: open-source AI agent hub you deploy on your own Vercel account — evilrabbit_ · 2026-11-10
- Preprint: Learning Tool-Output Projections from Agent Traces Cuts Billed Tokens up to 56% — solyarisoftware · 2026-11-10
- Dev says he hasn't manually written code in 2 years, dating back to Alex Sidebar work — rudrank · 2026-11-10
- Fable 5.2 built the same game twice—with and without ElevenLabs MCP, and the audio gap is striking — mhdfaran · 2026-11-10