Developer Slams New Claude Model as 'Stubborn' and Bad at Following Coding Instructions
sibidharan · reddit · 2026-08-04
A developer recently vented frustrations on Reddit about Anthropic's latest coding model (referred to in the post as Opus 5), describing a severely degraded coding experience. Despite explicit instructions, detailed project plans, and context provided in CLAUDE.md, the model reportedly ignores direct commands and stubbornly generates what it thinks is correct.
The author notes that while older versions might occasionally forget edge cases, this new update wastes time and tokens by arguing with the developer instead of following instructions. The user also criticized the model's poor UI/UX generation capabilities, highlighting a growing pain point for developers relying heavily on AI agents in complex, long-term projects.
Related event: Anthropic Faces Developer Backlash Over Declining Model Performance(11 posts)→
More from coding & agent
- He benchmarks new models by having them build their own copilot — $28 of compute — bucolucas · 2026-10-27
- Should agents be allowed to change plans mid-task? Reddit debates the drift boundary — wistfulcracker84 · 2026-10-27
- HeyGen + Canva MCP turns slide speaker notes into an avatar presentation video — HeyGen · 2026-10-27
- Local beats cloud: Qwen 3.8 FN handles private student data as well as frontier models — dr_lm · 2026-10-27
- Cloudflare Computer adds agent interruption recovery and approval pauses in v0.7.0 — irvinebroque · 2026-10-27
- LLM Tails cuts agent input tokens 80% by making agents read before they write — savadks1818 · 2026-10-27