Agent Scale Lessons: Shared Specs, Adversarial Pairs, Retiring Long-Run Agents
LukeW · x · 2026-11-21
Luke Wroblewski's notes from the Crazy Agent Scale event at Sutter Hill Ventures, where founders running agents at extreme scale shared what works:
- Clement Pang (Wavefront co-founder, open-source orchestration system Intent): give agents a spec living outside their context window in a shared space they can read, search, and edit — agents then self-organize around it for specs, logs, research, and UI mockups;
- Intent represents work as a task graph; since models often stop early thinking they're done, it quietly hints when finished tasks unlock new ones, driving work to completion;
- Organize agents like a company of specialists via system prompts; a useful pattern is adversarial pairs — one PM agent interrogates the user to build a spec while a second only critiques it;
- No context can be a feature: code reviewers who never saw the implementation discussion regularly spot a third option;
- Long-running agents can "retire" to avoid context compaction, with a fresh agent taking over.
More from coding & agent
- Agent-scale night: 21-person company hits $100M ARR with zero support teams — LukeW · 2026-11-21
- "Diagram Driven Development": make the AI-drawn logic diagram the single source of truth — max__drake · 2026-11-21
- shadcn launches UI SDK: reusable component behaviors designed for agent-built interfaces — shadcn · 2026-11-21
- Claude Code dual-model setup: one model plans and reviews, subagents write the code — PrajwalTomar_ · 2026-11-21
- Dev argues agent approval flows should let you approve actions by voice — NYCounihan · 2026-11-21
- ENZO v1.7.0: one-click self-hosted AI workspace with evidence-citing research mode — anonymous20256 · 2026-11-21