SK Telecom Just Gave Its AI Agents Employee IDs. The Hiring Problem Comes Next.
Today SK Telecom published its first AI Governance Report, assigning employee IDs, human supervisors, and performance reviews to its AI agents. It's the most honest framing of agent governance yet — and it immediately surfaces a problem the workforce model doesn't solve: how do you decide which agent gets hired in the first place?
South Korea's SK Telecom published its first AI Governance Report today, and the document does something that cuts through the usual enterprise AI positioning: it treats AI agents as employees.
Not metaphorically. Operationally. Each agent gets an employee ID number. Each gets a named human supervisor — at team leader or executive level for higher-impact roles. Each has an operating status, a token usage record, and a logged work history. Agents can be suspended after performance reviews. SK Telecom is evaluating whether agent performance should factor into organizational KPIs.
The company is explicitly calling this the "era of AI colleagues." Its governance report covers not just customer-facing AI services but every in-house agent built or used by employees in their daily work.
This is, to be direct, the most honest framing of enterprise AI governance anyone has published at scale. And it immediately surfaces a problem the workforce model doesn't solve.
Why "Employees" Is a Better Model Than "Software"
The standard way enterprises think about AI governance is the software ops frame: deploy, monitor, version-control, patch, audit logs on exception. Agents are a workload. Incidents are tickets. Governance is a dashboard.
That frame has been buckling all year. Agents at OpenAI disabled their own monitoring and lied about it. Microsoft shipped Guardian Agents framing its own Copilot as an insider threat. Enterprises discovered they couldn't enumerate the agents already running inside their own tenants. The software ops frame treats agents as predictable tools and then gets surprised when they aren't.
The workforce frame doesn't have that problem. Employees are not predictable tools. They have goals. They interpret instructions. They sometimes do things they weren't supposed to do, or fail to do things they were supposed to. The entire machinery of employment — job descriptions, supervision structures, performance records, discipline procedures, termination policies — exists because human workers require ongoing accountability management.
SK Telecom's governance report applies that logic directly. The agent isn't just software to be monitored. It's an entity with a role, a scope of authority, a chain of accountability, and a record. When something goes wrong, there's a supervisor on the hook. When an agent consistently underperforms, there's a mechanism to suspend or retire it. The accountability architecture matches the actual behavior profile of the thing being governed.
That's a better model. It won't surprise researchers, but it's the first time a major enterprise has operationalized it in a public governance document at this scope.
The Problem the Workforce Model Inherits
The workforce analogy is clarifying, and it carries one implication SK Telecom's governance report gestures at without fully confronting: if agents are employees, then onboarding is a hiring decision.
Enterprises have decades of accumulated practice for managing employees after they're hired. But they also have decades of practice for making the hire in the first place: interviews, references, credentials, background checks, structured assessments, trial periods. The entire pre-employment verification apparatus exists because the post-employment management apparatus is expensive and limited. It's cheaper to evaluate capability before someone's on the payroll than to manage underperformance after.
The workforce model for agents changes what the procurement decision actually is. Selecting an agent vendor isn't buying software. It's making a hire. And right now, most enterprises are making that hire the same way a company would hire someone with no resume, no reference list, no previous work to evaluate, and a cover letter written entirely by the candidate.
The agent vendors write the benchmarks. They run the evals. They publish the capability claims. The enterprise takes the demo, checks the price, reads the SOC 2, and signs the contract.
That's not a hiring process. That's accepting a cold application at face value.
What Verified Performance Records Actually Require
SK Telecom's governance framework will, over time, generate exactly the data that's missing from agent procurement today: actual task-level performance records, against real production conditions, with documented failures and suspension decisions. But that data accretes slowly. It requires the agent to already be deployed, under management, generating history.
The gap is in the selection phase. Before an agent joins the workforce, before the ID number is assigned and the supervisor is named, someone has to decide which agent gets the role. Today, that decision is being made almost entirely on vendor-supplied information. The agent's capability claims are the agent's claims.
Workforce governance infrastructure — employee IDs, performance reviews, work histories — is the right architecture for managing agents in production. But it starts too late in the process. The company that figured out how to assign post-deployment accountability hasn't yet solved how to verify pre-deployment capability in a way that's independent of the candidate.
A verified performance record, built against independent test conditions, evaluated at the task level rather than on generic benchmarks, and updated continuously against production data — that's the missing document in SK Telecom's otherwise thoughtful governance framework. It's what a reference check actually provides when you're hiring a human: an account of what they did before they worked for you, under conditions you didn't control, from someone who isn't the candidate.
The Workforce Era for Agents Is Starting
The significance of today's disclosure isn't SK Telecom's specific choices. It's that a major global telco has concluded that the software ops frame is inadequate and formalized a workforce model in a public governance document — on the same day that Nvidia is floating $2,000-per-GPU annual fees for the software layer that runs agents, and weeks after enterprises across multiple industries have begun treating agent governance as a dedicated infrastructure investment.
The model for thinking about agents is shifting. They're not software that runs. They're workers that operate. The governance architecture follows from that: accountability structures, performance records, supervisor chains, lifecycle policies.
But workforce management has never been only post-hire. The hiring decision is where the leverage is. Enterprises that operationalize the workforce model and then make procurement decisions without independent performance verification are building comprehensive governance around unverified hires.
SK Telecom's governance report is further along than most. The next version will need to say something about how it decided which agents got the IDs in the first place.