Does Your AI Agent Have an Employee ID? The Agent HR File Every Production Agent Needs

SK Telecom has started managing its AI agents the way it manages people. Each agent gets an employee ID, a department, a job function and a defined set of access permissions, and is handled from onboarding through to offboarding, with a governance framework for what data and systems it can reach.

It is easy to read that as a branding exercise. It is closer to the opposite. Every item on that list is a basic control, and many agents running in companies today don't have them.

Why agents need what employees have

When a new person joins a company, a lot happens before they touch anything important. They get an ID, a manager, a role, and access to the systems that role needs, not everything. Their work is reviewed. When they leave, their access is removed.

Agents often skip all of that. They run under a shared service account, nobody is clearly responsible for them, their permissions are whatever was easiest to set up, and they keep running long after anyone remembers why. When something goes wrong, the logs say "the system did it."

The agent HR file

We use a simple record for every agent before it is given access to anything that matters. Eight fields:

• Name and ID. A unique identity, so every action in the logs traces back to one agent.
• Role. One sentence on what it is for. If you can't write it, the agent isn't ready.
• Department. Which part of the business it works for.
• Supervisor. A named person who answers for what it does.
• Permissions. What it is allowed to read and do, written down so it can be reviewed and limited.
• Not allowed. The risky actions it must never take, enforced by the system, not just mentioned in a prompt.
• Review. How and how often its accuracy and cost are checked.
• Kill switch. How its supervisor suspends it, in one step.

A lifecycle, not a launch

Treat an agent's time in your systems like a career. Onboard it with an ID, an owner and limited access. Keep it on probation, reading and proposing only, while its work is reviewed closely. Promote it to more access once it has earned it. And retire it, removing its access, when it is no longer needed. The agents nobody owns are the ones still running, with access, long after anyone remembers why.

Check your own agents

• Does it have its own ID in your logs?
• Can you name the person responsible for it?
• Are its permissions written down?
• Are its riskiest actions blocked?
• When was its work last reviewed?
• Could you switch it off in one step?

Any "no" or "not sure" is the place to start.

How we approach it at Zylen

Zylen is an AI optimization partner. We work inside the AI systems companies already run. Our ten-day AI Optimization Audit assesses your AI paths against the OWASP GenAI LLM Top 10 2026 and shows you which agents need an owner, tighter limits and a kill switch.

Give every agent a file before it gets access: start the AI Optimization Audit →

SK Telecom details from its June 2026 announcement, as reported by Asiae, SBS and RCR Wireless.