Every AI agent now gets an ID. None of them gets a job.
Microsoft now gives every new Copilot Studio agent its own identity and a human sponsor. An identity says who an agent is, not what work it owns.
Agents are becoming employees on paper. Each one gets an identity, permissions, an audit trail and a named person responsible for it. That is progress, and it is the easy half. An ID tells you who the agent is. It does not tell you what work the agent owns, who hands it that work, or what "done" means. Companies that stop at the badge will have a tidy directory of agents and the same operation as before.
What Microsoft changed
According to Microsoft's Copilot Studio documentation, after May 2026 Copilot Studio automatically creates a Microsoft Entra Agent ID for each new agent, and administrators can no longer opt out. The identity makes the agent's connector permissions visible to admins, lets them apply Conditional Access policies, and brings the agent into audit logging and lifecycle management. The agent's owner is added as a sponsor, which Microsoft describes as giving governance traceability, accountability for the agent's lifecycle and alignment with organizational ownership policies.
Read that as a design statement: an agent is a thing that should have a responsible person. Good. Notice what the identity layer cannot contain, though. It records who may access what. It has no field for which outcome this agent is accountable for.
A badge is not a job description
When you hire a person, access is the last step of onboarding, not the first. Before it come the questions that make the work possible:
- What is the role? Which stage of which workflow is theirs, and where does it end?
- Who hands work over, and to whom does it go next?
- How is it measured? One number, visible to the person who owns it.
- When does the role change or end?
Agents skip all of this because the platform lets them. You can create one in an afternoon, it gets its identity automatically, and it starts producing activity. Activity looks like work. In a system nobody redesigned, it is mostly motion.
The sponsor problem
A sponsor is the right idea, and it fails in a predictable way. If the sponsor is whoever built the agent, usually someone in IT or an innovation team, they can answer "what can it access?" but not "is the queue shorter?". The person who can answer that runs the operation: the head of customer service, of reporting, of fulfilment.
The sponsor of an agent should be the owner of the workflow it sits in, not the person who configured it. If that person does not exist, the finding is not about the agent. It is that the workflow has no owner, and no agent can fix that.
- 01Define the workflowStart, end, what counts as done
- 02Name the ownerSomeone in the operation
- 03Set the one numberVisible from day one
- 04Then grant accessIdentity and permissions
Identity is the fourth step, not the first.
What the evidence looks like
The cases we can point to start from the operation, and the autonomous parts come after. At Rodo, WhatsApp, social and web enquiries were brought into a service operation with a measurable resolution path. In the first measured window, 82% of live-created enquiries were resolved (cohort 31 August–24 September, from the Rodo operational report of 24 September 2026; figures describe pipeline records, not all conversations). It was a first measured window with no comparable earlier baseline.
The point is not the number. It is that every enquiry had a place, a stage and an outcome before anything was automated. An agent working inside that path has a job because the path exists.
A test for this week
List the agents running in your company, from the pilot nobody remembers to the one the CEO demoed. For each, write one line: this agent owns [stage] of [workflow], its owner is [name], and we know it works when [number] moves.
Every agent you cannot finish that sentence for does not have a job. Fix the sentence before you review its permissions.