Most companies talk about AI as software they deploy. SK Telecom has taken a different position entirely: AI agents are digital employees, and the company is managing them accordingly.
The carrier’s AX Innovation 2.0 program assigns each AI agent an employee-style ID, a department, a defined job function, and role-based access to systems and data matched to that function. Agents are brought on through a process described as similar to human hiring, and they are managed through a full lifecycle that ends with formal termination when the agent’s role is complete. That language is doing real work, not just branding work.
In practice, each agent’s scope mirrors the job description of a human employee in a comparable role. A marketing agent should have the same data access as a human in a comparable marketing role, not more. A network-operations agent has a different permission set entirely. The intent is that the boundaries are structural, not just policy.
The structural goal is one AI agent per employee across 25 companies and 80,000 workers. Employees who are not developers can build their own agents using no-code tools and share them through an internal Agent Store. The platform is rolling out across SK Group by year-end.
Why the Framing Matters
When you treat AI agents as software, governance is an IT problem. You set API limits, monitor endpoints, and audit logs. When you treat AI agents as employees, governance becomes an organizational design problem. Who authorizes the agent? What data can it access? What happens when it makes an error? Who is responsible? Those questions require answers at the business level, not just the technology level.
SK Telecom is also applying a Zero Trust security model to agent interactions, which is the right call. An agent that can access every system by default is an attack surface. An agent with role-scoped access that matches its defined function is a manageable risk.
The one-agent-per-employee ambition is the most striking part. That is not augmentation. That is organizational redesign at scale. It implies that every role in the company will eventually have a defined AI counterpart with its own identity, permissions, and accountability structure. That is a fundamentally different way of thinking about how work gets done. And at 80,000 people across 25 companies, it is not a pilot.
Source: RCR Wireless, June 18, 2026
Get new articles from Robin Green delivered directly.
Insights on AI leadership, the future of work, and human collaboration. No noise. Unsubscribe any time.
Subscribe free
Interesting, Robin. The pattern I keep hearing inside the pros and cons is the part nobody planned for: the quiet hiring back of people after the automation overshoots, or once the longer-term results come due.
I sit in the augmentation and integration camp. Companies that only automate eventually hit a wall, or will get outcompeted by the ones that kept human judgment in the loop. #SmartIsTheAnte, per Paul Gibbons.
Here is the gap in the one-agent-per-employee model. Every agent performs a function. None of them reads the human and organizational layer those functions actually run on. Blind spots, where trust breaks, which changes survive contact with the floor, where innovation actually delivers a better result. That is judgment intelligence, attention to behavior rather than throughput, and it is usually what decides whether the redesign holds.
That gap is one area we are building iVASA toward. An agent whose job is the particular individual and organizational behavior itself, not the task sitting on top of it. SK Telecom just made the case for why that hire belongs on the org chart.
@bethmacdonald is one leader in this space focused squarely on this gap.
Dan, the emerging rehiring pattern is one of the most honest signals in all of this. When the automation ROI math gets recalculated two years out and the human layer turns out to have been load-bearing, there is no press release. The headcount just creeps back.
The gap you’re describing is real. Most deployments optimize function performance. None of them map what the function runs on: where trust is thin, where the change narrative broke, which managers are actually moving work and which ones are performing compliance. That is not a data problem. It is an attention problem, and it does not show up in throughput metrics until it is already too late.
Worth watching whether SKT’s org hire model starts closing that layer or just adds another function to the stack.