AIBRIEF
After a May rollout to 'thousands' of non engineering staff, the open source release is meant to fix static apps and Model Context Protocol (MCP) collaboration gaps.
Cloudflare open-sourced the second version of its internal agent platform on Wednesday, publishing the same code it rolled out to every employee in May. The release, branded "Cloudflare OS" but better described as an agent workspace layer tied to company-specific context, tools, and a shared skill library, is positioned as a step toward letting any organization do the same.
In a blog post, Cloudflare says "thousands of people across every function, many of them outside of engineering" used v1 daily to write documents, automate tasks, and build small internal apps. The post is candid about v1's limits: apps were static rather than live software connected to internal systems, deterministic jobs still required re-running an agent skill, and collaboration exposed a fundamental challenge around MCP (Model Context Protocol) server access, the question of which tools get exposed to which agents.
A Hacker News thread reframes the platform as a remake of Sandstorm.io on Cloudflare Workers, with per-document sandboxes that one commenter claims are strong enough that "the AI cannot introduce a significant security bug." That claim is unverified; the usage figures are self-reported; and the CIO blog post that would carry the internal-deployment narrative is not in the source bundle.
The open question: whether sandboxing plus a shared skill library solves the org-wide-agent problem, or just offers the first vendor's answer to it.