AI Agents: Authority Not Just Ability

The practical issue with artificial intelligence agents isn't just how clever they are, but what they're allowed to do. Jonathan Harris explains why governance needs to precede deployment, moving beyond benchmark scores to focus on precise control over access and actions. This week, we examine how the hype around agentic AI often distracts from fundamental questions of authority and security. We also look at the layers of engineering behind AI systems – prompt, loop, and graph – and why confusing them leads to costly misunderstandings. From industrial safety to the structure of the web, the real progress lies in architecture and careful control, not just more confident claims or faster models. Scepticism, as Jonathan Harris notes, is basic maintenance.