Your Agents Need a Manager, Not Just an Admin
AI agents are getting employee IDs, named supervisors, performance reviews and retirement dates. That's workforce management. Workflow practitioners should own the design before it defaults to whoever holds the admin console.
In the space of a few weeks, a major telecom operator and three of the largest enterprise software vendors landed on the same idea. SK Telecom gave its AI agents employee ID numbers. IBM shipped a registry that records each agent's owner and lifecycle status. Microsoft is giving agents their own Microsoft 365 accounts. Workday is buying a company that manages access for humans and agents in one identity layer.
Read separately, these look like security and identity announcements. Read together, they describe something else: agents are being managed the way organizations manage people. They get hired, assigned a boss, evaluated and eventually let go.
That raises a question most organizations haven't asked yet. When an agent needs a manager, who designs what that manager does?
What moved in October
Date | Who | What changed for agents |
Oct 27 | Agent Identity, with HashiCorp Vault, registers agents in a central inventory and records each one's owner, lifecycle status and granted authority. It also classifies risk and documents approved tool connections. | |
Oct 26 | First AI Governance Report. Agents get employee ID numbers and are recognized as an official new workforce. High-autonomy agents need a team leader or executive as supervisor. Low-value agents are suspended or retired. | |
Oct 26 | Microsoft (via Big Hat Group) | Copilot Studio agents can receive Microsoft 365 accounts, be mentioned in email and Word comments, and run from schedules or signals. |
Oct 13 | Agreed to acquire C1.ai, which manages access for employees and agents, and fold it into Workday's Agent System of Record. |
The scale behind this is the real story. IBM's announcement cites a Gartner prediction that the average Fortune 500 company will run more than 150,000 agents by 2028, up from fewer than 15 in 2025. No organization manages 150,000 of anything by hand.
The pattern: identity, owner, review, retirement
Strip away the product names and every one of these moves maps to something organizations already do for people.
Workforce practice | The agent version | Who is doing it |
Hiring and an employee ID | A registered identity and account | SK Telecom, IBM, Microsoft, Workday |
A reporting line | A named owner or supervisor | SK Telecom, IBM |
Role-based access | Permissions granted and revoked by task | IBM, Workday |
Performance reviews | Regular evaluation tied to organizational performance | SK Telecom, Microsoft |
Offboarding | Suspension, retirement, decommissioning | SK Telecom, Microsoft |
SK Telecom goes furthest. It is weighing whether to share agents' performance with employees and, longer term, reflect it in organizational KPIs. That is the language of a workforce plan, not an IT asset register.
Why this is management, not administration
An admin answers "can this agent access that system?" A manager answers "should this agent be doing this work at all, and is it doing it well?" The tools announced this month are excellent at the first question. They are silent on the second.
An identity registry can tell you who owns an agent. It can't tell you what that owner is supposed to do on a Tuesday. It can record that an agent is active. It can't tell you whether the workflow it sits in got faster, or just noisier. It can revoke credentials. It can't decide which agent should hand off to which human when an exception lands.
As one analyst put it in Techstrong's Workday coverage, an agent "cannot fix conflicting employee records or decide which policy is correct." Someone still has to. That someone is a manager, and the job is designed in the workflow, not the console.
The risk is drift by default. If nobody designs the supervisor role, it lands on whoever provisioned the agent. Usually that's IT or the person who built it. Neither is positioned to judge whether the agent is creating value in the work around it.
The agent supervisor design questions
SK Telecom's rule that high-autonomy agents need a team leader or executive as supervisor is a good start. It names who. The harder part is what they do. These are the questions a workflow practitioner should answer for every agent that touches real work:
What work is delegated, and what isn't? Which steps does the agent own outright, which does it draft for human approval, and which stay human? This is explicit delegation, and it belongs in the workflow design, not a permissions screen.
What does the agent hand off, and to whom? Every agent sits between upstream and downstream steps. The handoffs, including exceptions, need a named human receiver.
What does "good" look like? Token usage and uptime are activity measures. The supervisor needs workflow measures: cycle time, rework, exceptions raised, approvals rejected.
How do we know it's drifting? Models change, data changes, policies change. Someone has to watch whether the agent still matches the references it was built on. SK Telecom commits to ongoing monitoring after launch for exactly this reason.
When does it retire? Retirement criteria should be set at launch, not discovered when costs spike.
WMI's AI Workflow Governance framework names three of these directly: Explicit Delegation, Reference Alignment and Drift Detection. The IDEAS Workflow Ownership Model gives each agent's supervisor a clear accountability loop to run.
What workflow practitioners should claim now
The defaults are being set this quarter, inside identity platforms and HR systems. Practitioners who wait will inherit them. Four moves are worth making now:
Get in the room with IT and HR. Identity teams own the registry. HR is starting to own the workforce framing. Neither owns the workflow. Offer to supply the part they can't.
Add a workflow field to every agent record. Owner and lifecycle status aren't enough. Each registered agent should name the workflow it serves, the step it owns and the human it hands off to.
Define the supervisor role before the first review cycle. Write down what an agent supervisor checks, how often and against which measures. One page is enough to start.
Map before you multiply. Workday's own product lead warned about the trap of too many agents. A current-state map of the workflow shows where an agent belongs, and where it would just add another handoff.
The real signal
When a telecom operator and three enterprise vendors start describing software with words like employee ID, supervisor and retired, the category has shifted. Agents are joining the workforce. The organizations that get value from them will be the ones that design how they're managed, not just how they're logged in.
The console can tell you an agent exists. Only a well-designed workflow can tell you whether it should.
Sources
SK Telecom issues first AI governance report, Seoul Economic Daily, Oct 26, 2026
IBM aims to bring AI agent sprawl under enterprise control, Network World, Oct 27, 2026
Copilot Weekly, Oct 19–26, Big Hat Group, Oct 26, 2026
Workday unveils agentic AI tools and Anthropic partnership to automate HR, finance workflows, Techstrong.ai, Oct 13, 2026




Comments