Google releases open-source Scion for multi-agent AI teamwork
Google’s open-source Scion framework lets developers run multiple AI agents—like Claude Code, Codex, or Gemini CLI—as a coordinated team rather than competing models. Released under GoogleCloudPlatform on GitHub, Scion enables agents to split tasks (e.g., requirements, coding, testing, review) across independent processes, sharing files and verifying outputs at each stage.
Key points
- Scion lets developers run Claude Code, Codex, Gemini CLI, and others as a team with shared files and verification steps
- Google’s film production experiments used Scion to create 25 projects totaling 44 minutes of footage, but found agents sometimes failed to complete tasks
- File-based collaboration outperformed message-based, as files persisted when agents crashed or restarted
The project emerged from Google Cloud’s experiments with 10 AI film production teams, which used Scion to divide workflows into seven stages with verification gates. Teams created over 25 productions totaling 44 minutes of footage, but also uncovered flaws: agents sometimes produced incomplete work (like a 94-byte placeholder) or lost context when restarted. Google found that file-based collaboration (e.g., Git worktrees) worked better than message-based communication, as files persisted even if agents crashed.
Scion supports local CLI, Workstation UI, and containerized setups, with authentication handled per model provider. It’s not an official Google product but a pre-1.0, rapidly evolving project. Installation via Homebrew is recommended, though costs for API usage and compute scale with agent count.
Claude Code, Codex, and Gemini CLI can all be on the same AI team. The OSS 'Scion' is interesting
note.com · 14 October 2026
Loading the full article…
This text was published by note.com and written by KitWorks. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Agents & Tools
All →- OpenAI introduces Dots agent for asynchronous task delegation · 2 src
- OpenAI and Anthropic outline safety frameworks for financial agents · 1 src
- Anthropic adds OpenTelemetry trace export to Claude Platform for debugging and monitoring · 2 src
- The reason AI 'cannot see internal company information' is that it is not connected — MCP is a common port for connecting Claude to external tools · 1 src
- Linus Torvalds uses Anthropic's Claude Code to draft Linux kernel changes · 1 src
Comments
via GitHub Discussions