SKT Gives AI Agents Employee IDs, Then Reviews How They Perform

The ID goes to agents that pass a review

Once an agent is built it is registered, and if it clears a separate internal review it gets an employee ID. From then on its tasks, owner, department and operating status are managed together.

It is the same idea as a name badge: someone is clearly responsible for each agent.

The more autonomous the agent, the more humans watch

Who owns an agent depends on its impact and autonomy. A personal assistant can be managed by the person using it, while a high-impact or highly autonomous agent answers to a team leader or executive. Functions and searchable data are limited to the task, and more autonomy means more human review and approval.

An 'Agent OS' logs each agent's runs, usage, change history and the models and tools it used, and can halt an agent if something looks wrong.

Weak results can end an agent's run

ID-holding agents are assessed on task completeness, impact scope, data access and contribution, and the outcome decides whether to keep, improve or retire them. SKT is also considering tying the results to team KPIs in the long run.

Three steps through 2028

This year is groundwork: pre-launch checks, identification and owner assignment, access standards. 2027 brings an integrated system with tiered permissions and stronger human review, and 2028 aims for lifecycle management from operation to evaluation and retirement.

The ISO certification now covers internal agents

SKT won ISO/IEC 42001 certification for its customer-facing AI services in 2024, a first among Korean telecoms, and in October extended it to the management system for internal AI agents.

Chief Privacy Officer Cha Ho-beom said AI governance matters more in the era of AI agents and that SKT will strengthen its 'AI trust' management. The report describes the company's own framework; it is not an independent audit of results.