44% of CEOs can’t yet show a return on AI. This week, what Anthropic’s agent-heavy workforce, Barclays’ three-year rebuild and a new wave of agent workspaces tell us about where the return actually comes from.

Ontologies are becoming the standard way to give AI agents shared business context. UK AI policy puts AI security on the board agenda. And AI-written code is changing what architects need to review.

More enterprise security teams now have access to advanced AI tools. Ecuador is extending its use of Palantir to tackle money laundering. And ChatGPT has started building interactive interfaces, not just text answers.

Microsoft Copilot Autopilot and OpenAI dots push agents further into everyday work. That raises the stakes on agent identity, memory and oversight. AWS has open-sourced an ontology tool, and OpenAI has paused Astra 6.1.

Legal AI is only as good as the proprietary knowledge behind it. Frequent model changes make evaluation a recurring cost. Typed decision engines are challenging the default of using an LLM for everything. And governance is the foundation of any company brain.

The case for pacing frontier AI is now as much economic as ethical. Microsoft’s Humanist AI code of conduct draws a line on machine consciousness. And enterprise buyers are growing reliant on a small number of model and infrastructure vendors.

OpenAI’s Astra reasons internally, which makes its decisions harder to inspect. Five Eyes security agencies have published guidance for governing AI agents. Multiplayer AI points to agents that serve teams rather than individuals. And Mistral’s €3bn raise makes sovereignty a buying criterion.

Nvidia’s deal with Hugging Face reshapes the open-model ecosystem. AI is changing how consulting work gets delivered. Document extraction and agentic software delivery practices are maturing. And EU search rules are reshaping enterprise AI.