How should IT prepare for staff using their own AI agents at work?

Personal AI agents are different from chatbots because they act inside real systems, often using a member of staff's own login, so IT needs to plan for what they can change rather than only what data they see. Blocking every agent rarely lasts, because people move to personal devices and IT loses sight of the activity. A better approach is to find out what is already in use, agree a short set of categories, and control the consent that lets an agent reach email, calendars and files. Every agent that touches shared work should have a named human owner who answers for what it does. Agent actions should also be visible to normal change and incident processes, since an untracked change made by software causes the same Monday morning surprises as one made by a person. John Sansbury's guiding principle of optimise and automate applies here: fix the underlying process first, or an agent will simply make a messy process fail faster. Finally, offer a supported option in the service catalogue so that the safe route is also the easy one. The article includes a worked example of a small firm that reviewed app consents, kept two approved uses, named an owner and added an incident category, and it points to the Ownership and boundaries canvas in the ESMi AI Enablement Toolkit as a practical next step.