All insights

Weekly23 October 20261 min read

This week in AI: open standards for agents

Anthropic has donated Agent Skills to the Agentic AI Foundation, joining the Model Context Protocol. Shared standards make it easier to switch tools and keep what you have built.

What happened

On 22 October, Anthropic donated Agent Skills to the Agentic AI Foundation. Skills are a way to package instructions, scripts and reference material that an AI agent loads when a task needs them. The foundation, part of the Linux Foundation, already looks after the Model Context Protocol, which connects AI tools to data and systems. Separately, Google released EmbeddingGemma 2, a small open model that represents text, code, images and audio in a single searchable form.

Our take

Open standards are not exciting, but they are what make technology safe to buy. The Technology Code of Practice asks public bodies to use open standards and avoid lock-in for good reason: a system built on a standard can move to another supplier or tool.

With Skills and MCP both governed in the open, two important parts of an AI system can now be portable. Your connections to systems (MCP) and your organisation's know-how written down for AI (Skills) can be built once and used with tools from different vendors.

We build on both. When we hand a service over, the skills and connectors are part of what the client owns, documented and ready to use with whichever model they choose next.

Keep reading

Related insights