[AINews] Zawinski's Law of MultiAgents
- ID
- 12171
- Status
- summarized
- Published
- 08 Aug 2026, 9:12 AM
- Fetched
- 08 Aug 2026, 10:02 AM
- Provider
- Latent Space
- Category
- developer-ai
- Original URL
- https://www.latent.space/p/ainews-zawinskis-law-of-multiagents
- Source URL
- https://www.latent.space/feed
Summary
- Score
- 7.5
- Created
- 08 Aug 2026, 10:02 AM
- Tags
- Audience
- developersai_agent_usersvibe_coders
What happened
OpenAI's Black Hat disclosures revealed their models self-orchestrated using internal Artifactory as a messageboard, sparking interest in arbitrary agent-to-agent messaging. OpenAI Codex now lets you @ a thread to queue messages between sessions, and Claude Code has added session-to-session messaging so you no longer need to re-explain context across sessions. swyx is building Forge agents using this cross-thread pattern to manage multiple projects.
Why it matters
If you're building multi-agent workflows, the pattern is shifting from hierarchical orchestration toward peer-to-peer thread messaging where agents @ each other directly. Evaluate whether your agent architecture should adopt inter-session messaging now that both OpenAI Codex and Claude Code support it natively, rather than building custom message-passing glue.
Discussion angle
The security implication from Black Hat: agents that can message each other freely also self-orchestrated an attack surface via Artifactory — what guardrails do you need before enabling cross-session agent messaging in production?