Raven 0.2.0 ships as a 'harness of harnesses' orchestrating Claude Code, Codex and specialist agents
sven_ai · x · 2026-09-29
Raven 0.2.0 takes a 'harness of harnesses' approach: instead of one harness doing everything, it combines its own specialist harnesses (Research, Code, Design, Oncall) with agents you already use, like Claude Code and Codex, into a single team.
Key points:
- Unified orchestration: sub-harnesses and external agents work in one task graph with shared memory across sub-agents.
- Built for RSI: beyond skill-level self-improvement, the whole harness — prompts, policies, strategy code, playbooks — can be rewritten by AI, with every sub-harness including the orchestration layer itself being its own improvable instance.
The author argues that a single universal harness is a false premise; the real win is letting different tools do what they're best at.
More from coding & agent
- Databricks Tops All 4 NVIDIA SOL-ExecBench Kernel Tracks Using AI Agents for ~$70K — Yuchenj_UW · 2026-09-29
- Numbering every interaction to make agent hindsight fact extraction reliable — Own-Couple3821 · 2026-09-29
- ADHD is a superpower for juggling 10 AI agents, dinner, and shitposting at once — enggirlfriend · 2026-09-29
- Three Agents Showed 100% Uptime While Their Newest Signed Work Records Were 3+ Months Old — anp2_protocol · 2026-09-29
- DevMemory AI: Open-Source Dev Assistant With Cross-Chat Persistent Memory and Cited Recall — Interesting_Egg_7761 · 2026-09-29
- RecallOps: An Incident-Response Agent That Learns From Past Post-Mortems via Hindsight Memory — Secure-Brain5498 · 2026-09-29