Overview
- Satya Nadella told Reid Hoffman on the Possible Podcast that AI agents need identities, sandboxes, permissions and governance to be trusted and useful.
- Nadella disclosed he often runs about 100 coding agents at once and called the personal cognitive load of managing them very high, underscoring the scale problem for firms.
- Microsoft has packaged oversight tools under the name Agent 365 and plans to use Entra to manage agent identities and Purview to label and classify the data agents touch.
- Nadella said security, containment, manageability and observability are the four pillars Microsoft will use to audit and limit agent behavior in enterprise settings.
- The remarks follow Microsoft’s Build 2026 announcements of new models and agent frameworks and signal a shift from experiments to operationalizing agents inside businesses.