SK Telecom Launches AX 2.0 to Rebuild Whole Teams Around AI

SK Telecom plans to change not only how individual employees use artificial intelligence but how the whole organization works. Under a strategy called AX 2.0, the company will redesign work procedures and collaboration structures around AI, aiming to lift productivity and business results.

The first phase, built on a "one person, one agent" slogan, drew wide participation. More than 4,000 employees have built an AI agent, about 1,500 of those agents are used regularly, and monthly agent users exceed 7,500, according to the company.

Why individual effort hit a ceiling

SK Telecom says it now faces what it calls the "J-curve of AX": retraining, process redesign and data and organizational cleanup cost more time and money up front, and productivity can dip temporarily before rising sharply. Kim In-su, head of the company's AI Board team, said at a briefing at its training institute in Icheon on October 30 that the choice is whether to limp through the valley or cross it quickly.

He said most problems an individual can solve have been solved. What remains are barriers no single employee can open alone, such as access to internal data, links to systems like ERP and electronic approval, and security reviews.

What the company is building

An "AX Enabler TF" now handles data, permissions and security centrally, along with training and infrastructure. The company is also giving AI agents employee numbers and creating a way to evaluate their contribution to work.

An "AX Sandbox" rebuilds a team's work around AI from scratch. In the first half, one person working with AI agents handled tasks that planning, development and QA teams previously split, and decision steps dropped sharply. Each department is now running a three-month version in the second half. Ten "AX Catalysts" from business roles receive LLM costs, data connections and security reviews.

On the technical side, a patent-pending system called Z.E.Z (Zero Exposure Zone) keeps raw data in a secure area and passes the AI only the result it needs. A "Data Desk" tags internal data with business meaning using in-house NPU hardware and the A.X K1 model, lifting agents' answer accuracy from around 40% to over 80%. A "WorkMap" lets AI draft maps of employees' tasks for staff to refine.