[Edaily Reporter Shin Yeong-bin ] A 50-meter-long reclaimer autonomously identifies and moves among piles of raw materials, while a unloader equipped with a 20- to 25-metric-ton grab shovels iron ore without any human operators. In the office, AI scans the company’s entire account system to detect suspicious transactions and even prepares financial risk reports.

POSCO DX(022100)The company is transforming into an “AI-native company” by leveraging Physical AI—which automates manufacturing equipment—and “AI employees” that perform specific tasks.

POSCO DX held a media day on the 29th at the Lotte Hotel in Sogong-dong, Seoul, and announced a business strategy centered on Physical AI and Agent-based AI as its two main pillars.

Choi Tae-hwan, Head of the Business Planning Division at POSCO DX, stated, “An AI-native company is not simply a company that has adopted AI, but one where everything is designed with AI as a given.” He added, “We will view AI not merely as a tool, but as an agent that performs tasks—essentially, as an employee—and we will redesign the entire business process accordingly.”

The strength highlighted by POSCO DX lies in its integrated capabilities spanning AI, Operational Technology (OT), and Information Technology (IT)—encompassing everything from sensors and equipment control on the manufacturing floor to process optimization, Manufacturing Execution Systems (MES), and supply chain and management systems. The company plans to validate its technology with the POSCO Group as its first client before expanding into external markets.

In its Physical AI business, the company has placed greater emphasis on the “robotization” of existing large-scale equipment rather than on humanoid robots themselves. This approach involves equipping cranes and raw material handling equipment with vision, decision-making, and control capabilities to enable them to operate autonomously or semi-autonomously.

Cho Seok-ju, Director of the AX Convergence Research Institute at POSCO DX, explained, “In process industries like POSCO, everything happens within the equipment,” adding, “While we cannot replace the equipment with robots, we are focused on robotizing the equipment to make it perform tasks similar to those of a robot.”

The company is also developing an “AI operator” that has learned the know-how of skilled operators. In this system, vision and sensors act as the operator’s eyes; a supervisor agent assesses the situation, and a control agent then issues commands to the actual equipment.

Director Cho said, “When highly skilled workers retire, it’s not just the people who disappear—the high-level tacit knowledge they possess disappears as well,” adding, “Initially, AI will assist operators, and in the long term, we are developing it to take the place of retired operators.”

Key projects include the automation of reclaimers in the raw material yard, the automation of plate cranes, and the automation of ship-to-shore raw material unloaders. Development of the reclaimer technology is largely complete, and it is currently undergoing trial operations. The plate crane adjusts the magnetic force of its electromagnets based on the shape and weight of the steel plates, while the ship-to-shore unloader aims to handle approximately 80% of the total unloading operations autonomously. The company plans to commercialize all three technologies in the second half of this year.

Robots have also been deployed for high-risk, high-intensity tasks such as removing impurities from the zinc-coating process furnace—which reaches temperatures of approximately 450 degrees—cutting cold-rolled coil strips, and transporting 65-metric-ton steel products. The scope of application is also expanding to include unmanned forklift logistics at the cathode material plant and hazardous material sampling operations.

Yoon Seok-jun, Head of the Robotics and Automation Center at POSCO DX, stated, “An intelligent factory is not a concept that aims for 100% automation by eliminating human workers,” adding, “It is intelligent, autonomous manufacturing where humans do what they do best, robots do what they do best, and AI does what it does best.”

Humanoids and Vision-Language-Action (VLA) models are still in the preliminary research phase. Tests are underway in a simulation space that replicates the actual steel mill environment in 3D, where humanoids are being tested operating test equipment. Director Yoon clarified, “We have not yet applied this to actual sites; we are currently in the stage of testing VLA models in the laboratory.”

In the administrative domain, the company highlighted “Agentee,” its in-house developed AI workforce platform. Agentee is not merely an AI agent that handles individual tasks; rather, it manages AI employees—who are assigned specific roles—from recruitment and evaluation to retraining, promotion, and retirement, taking a full lifecycle approach.

Kim Byung-seok, Team Leader of the POSCO DX AI Workforce Task Force (TF), emphasized, “What we need is not more AI agents, but AI employees who work together as members of the organization,” adding, “Automating individual tasks and driving organizational performance are two different matters.”

He continued, “In the future, the market will shift from companies that sell agents to companies that supply AI employees,” adding, “POSCO DX is not only focused on making AI operate more sophisticatedly but also on how to manage AI as an integral part of the organization.”

Currently, 113 types of AI employees are in operation within POSCO DX across finance, business management, human resources, and labor relations. Including group companies, more than 200 Agent AI systems have been developed, with over 500 executives and employees participating in more than 100 projects.

In the task of pre-closing risk checks, AI agents scan all accounts to identify irregular accounts and missing journal entries, then generate reports. A task that previously took about three hours has been reduced to around one minute. The company projects that applying AI agents to the entire accounting closing process will improve organizational productivity by at least 30%.

The company is also accelerating external commercialization efforts. POSCO DX aims to secure Agent’s first external contract by the end of this year and begin full-scale expansion starting next year. Physical AI plans to finalize its internal group technology in the second half of this year and then target industries with similar processes, such as steel, shipbuilding, and heavy industry.

Director Choi stated, “Our goal is to grow orders and revenue by approximately 30% within the next three years through our AX business, which encompasses both Physical AI and AgentAI.” POSCO DX plans to expand employees’ scopes of work by assigning new roles to those whose repetitive tasks have been reduced, rather than linking the adoption of AI to immediate workforce reductions.