30-Year Dev Veteran: My Agents Failed on Opus 5 Because My Workflow Was the Problem
antenore · reddit · 2026-10-10
A developer with 30 years in engineering, ops, security and architecture hit persistent agent failures—ignored briefs, regressions, basic bugs—while building an app on Opus 5. Weeks spent tuning hooks, memories and prompts brought no improvement even after Opus 5.5 shipped, despite widespread praise.
The realization: the workflow was the problem. After rewriting prompts, reviewing memories, and reimplementing hooks, everything worked—and even Opus 5 completed tasks accurately again. His takeaway: most of the time, the user's own setup is the root cause, not the model.
More from coding & agent
- Decisions API hits GA as agents automate cloud video editing via feedback loops — dkundel · 2026-10-23
- Swarms API changelog: 8 months of streaming, multi-agent orchestration, and dynamic model pricing — KyeGomezB · 2026-10-23
- A curated library of 50+ free AI agent resources: videos, repos, one Google Doc — ifioknkem · 2026-10-23
- Vercel adds no-account deploys: drag and drop a file to get a live URL — steren · 2026-10-23
- Dev runs 5 Claude Code sessions that coordinate via MCP Agent Mail to root his network cameras — doodlestein · 2026-10-23
- Vercel ships 'vercel deploy --temporary' for ephemeral agent-built deployments — cramforce · 2026-10-23