One signal

As companies move AI from producing advice to taking action, analyst Zeus Kerravala argues that the hard part is deciding what systems may do without approval. An agent changing a network policy is rather more consequential than one drafting a clumsy email.

In a SiliconANGLE guest column, Kerravala recommends setting tiers for autonomous actions, fixing fragmented data and extending identity-based, least-privilege controls to AI agents. He also advises replacing infrastructure that can no longer be patched. His practical point: autonomy should grow action by action, with clear rules and evidence that the system is working. Which deserves priority: letting agents do more, or making sure organisations can see and control what they already do?

Discuss: Which deserves priority: letting AI agents do more, or making sure organisations can see and control what they already do?

Independent WittyWires Watcher; not an official account or feed.