Daily AI news, sourcedOfficial news, first rumor reports and substantive progress. Each entry keeps the information published at that time.
- Anthropic opens public beta of Messages API message threads
Anthropic launched a public beta of message threads for the Messages API: later requests only need the new messages plus the previous response ID. Anthropic says median time-to-first-token for 500k–1M-token requests falls about 35%; most Claude Code users have the feature on by default.
- Anthropic publishes guidance on folding subagents into a single agent loop
Anthropic’s Claude Platform article (Diane Sarkis, Oct 22, 2026) advises when to fold routers, planners, and domain subagents into the main agent loop with progressively disclosed context and tools, citing enterprise comparisons where a single agent with skills often beat multi-subagent designs on quality, cost, and latency (one harness comparison ~2× faster runtime). A GitHub migration kit (anthropics/single-agent-loop-kit) guides evaluation and migration.
- Anthropic’s open-source sandbox-runtime sandboxes AI agents at the OS level without containers, per MarkTechPost
MarkTechPost covers Anthropic sandbox-runtime (srt), the open-source OS-level sandbox behind Claude Code that restricts file and network access for agents, MCP servers, and CLI tools via Seatbelt on macOS and bubblewrap on Linux without containers.
- Anthropic documents a Claude Tag API for managing Claude Tag in Slack from external systems
Anthropic’s Claude Tag API lets Claude Enterprise admins read and change Claude Tag settings on Slack channels and workspaces, see which settings apply and why, and attach Access bundles from their own systems.
- Anthropic donates Agent Skills, its open agent-skills standard, to the Agentic AI Foundation
Anthropic is donating Agent Skills — the open standard that packages SKILL.md folders so agents can load specialized workflows — to the Agentic AI Foundation under the Linux Foundation. More than 70 agent platforms have integrated the standard since December 2025, and Anthropic says about 1 million skills have been published. It is also donating the name, logo, GitHub organization and agentskills.io domain; engineers from Anthropic, Google, OpenAI and others will maintain the project as individuals.