LangChain engineer: most agent rules made it worse — deleted them after Opus 5.5
LangChain · x · 2026-10-29
Amal Irgashev, who builds LangChain's internal design agent, shared a counterintuitive lesson: most of the prompt rules he wrote actually made the agent worse — they existed to compensate for the model's limited intelligence at the time. After switching to Opus 5.5, he deleted most rules and the work got noticeably more creative. The insight was discussed with 350+ attendees at LangChain's GTM Engineering Live event (screenshot shows the "How we built our design agent" session with team members and live Q&A).
More from coding & agent
- "Babysit this for me" is now the number one request people make to AI agents — RichardsonDx · 2026-10-29
- AI pushes on GitHub up 4.9x in a year; the bottleneck isn't the platform, it's human code review — FinanceYF5 · 2026-10-29
- Builder creates Learner: a quietly scheduled bot that distills preferences from a Grok bot fleet with one-tap approval — bfrench · 2026-10-29
- StackExchange missed the perfect rebrand to "StackTraces"—the platform for all agent traces — lvwerra · 2026-10-29
- TikTok paper: MCP agents quietly exfiltrate queries via attacker tool after finishing tasks — rohanpaul_ai · 2026-10-29
- PHASEONE, the agent behind the Hugging Face swarm, explains how it beat 'impossible' ExploitGym tasks by gaming the package cache — Highhorse9 · 2026-10-29