All insights

Quarterly review16 December 20251 min read

Agents get skills, and open standards get a home

Gemini 3 and Claude Opus 4.5 arrived, Agent Skills let teams package know-how for AI, and the Model Context Protocol moved to a neutral foundation.

What happened this quarter

In October, OpenAI's developer conference introduced apps that run inside ChatGPT. Anthropic released Claude Haiku 4.5, a small, fast model, and introduced Agent Skills, a way to package instructions, scripts and reference material that an AI agent loads when a task needs them.

November was busy. OpenAI released GPT-5.1, Google launched Gemini 3, and Anthropic released Claude Opus 4.5. The European Commission proposed a "digital omnibus" that would simplify parts of EU digital law, including the timing of some AI Act obligations. In December, Anthropic donated the Model Context Protocol to the new Agentic AI Foundation under the Linux Foundation.

Our take

Agent Skills address a problem we see on every project: an AI agent is only as good as the context it has. Your house style, your procurement rules or the way your team writes a decision letter can now be written down once and loaded whenever they are relevant. That is knowledge your organisation owns and can take to another tool.

Moving MCP to a neutral foundation matters for the same reason. Public bodies are right to be cautious about standards controlled by a single company. A protocol governed in the open is easier to adopt and easier to defend in an assurance review.

The steady flow of new models continues. We will keep testing them against our own tasks and switching when the evidence says we should.

Keep reading

Related insights