Shepherd: Programmable meta-agents with reversible execution traces
shi_weiyan · x · 2026-07-30
Shepherd is a programmable meta-agent framework that enables reversible agentic execution traces, allowing observation, forking, rollback, modification, and resumption. It maps to algebraic effects in functional programming: declare the effect first, let the handler decide. Experiments show runtime supervisor improves pair-coding pass rate, meta-optimizer beats baselines on LiveCodeBench, and training meta-agent outperforms GRPO on Terminal-Bench 2.0.
Related event: Stanford Open-Sources Shepherd Framework for Reversible AI Agents(4 posts)→
More from coding & agent
- Why company AI agents go stale: adoption needs in-house knowledge, not just better agents — EdenEmarco177 · 2026-10-16
- Cheng Lou open-sources Handwired: hand-written weights turn code into training-free neural nets — dotey · 2026-10-16
- Building a portable context vault in Notion so every AI agent knows you from day one — ivanhzhao · 2026-10-16
- Lovable apps can now talk to customers on WhatsApp with inbox, reminders and templates — AlexandrePesant · 2026-10-16
- FDE-Gym benchmark: best agent fully solves only 14 of 95 real customer-delivery tasks — dair_ai · 2026-10-16
- Deep Research API showdown: OpenAI, Perplexity, Parallel, Tavily, Gemini, Valyu compared — bar_raiser333 · 2026-10-16