POSCO DX is pushing "robotization," which makes equipment move autonomously like robots, along with manufacturing-specialized artificial intelligence (AI) trained on the experience and knowledge of highly skilled workers nearing retirement. The plan is to build an "intelligent factory" where AI makes its own judgments and selects optimal tasks, accelerating AI transition at steelworks.
POSCO DX said at a press briefing held on the 29th at Lotte Hotel Seoul that it will pivot to an "AI-native company," with physical artificial intelligence (AI) for industrial sites and agentic AI for office operations as its two growth pillars. It plans to control industrial sites such as steelworks with physical AI, and fully introduce AI agents in areas like finance and human resources to improve performance.
Physical AI will be developed and introduced with a focus on building an "intelligent factory" where people, AI, and robots work together. Evolving from a smart factory that automates repetitive tasks, the intelligent factory is a smart plant where people, AI, and robots split up the work they each do best.
First, it will bring physical AI to industrial sites to enable "equipment robotization," in which large industrial equipment makes its own judgments and control decisions. By robotizing steelworks equipment that weighs tens of tons—such as large cranes that move steel and reclaimers that transfer raw materials in yards (worksites), as well as ship unloader equipment—the company aims to secure worker safety and boost productivity.
Cho Seok-ju, head of the AX Convergence Research Institute at POSCO DX, said, "These three items are currently in test operation, with commercialization targeted for the second half of this year." For the ship unloader, the goal is to make 80% of tasks that operators previously performed directly unmanned.
It is also building a system that converts the tacit knowledge held by highly skilled operators into core data for physical AI so that AI collaborates with people in onsite control rooms to operate robots and equipment. As part of that effort, it is developing an "AI operator" to take over the judgment and control tasks that people used to perform in onsite control rooms.
Cho said, "One of the challenges facing manufacturing today is that many highly skilled workers are about to retire," adding, "With retirement comes the problem of their high-level tacit knowledge disappearing."
He noted, "It is hard to find young people for high-intensity work, and processing costs for raw materials continue to rise, so we began looking for ways to codify skilled workers' expertise, experience, and know-how and to automate and autonomize repetitive tasks," presenting manufacturing AI and industrial robots as the solution.
Industrial robots will be applied to high-intensity, high-risk sites. Examples include a robot that removes impurities from the molten zinc pot in the galvanizing process at a steelworks, a band cutter robot in the cold-rolling process, and an autonomous guided vehicle (AGV) that transports 65-ton steel products. To support this, the company will design optimal robot systems based on simulation and develop a vision-language-action (VLA) model for mobile robots, including humanoids.
In office operations, it will apply "Aigenty," its in-house AI agent management platform. For the monthly recurring task of "pre-close risk checks," the agent automatically scans all accounts to find abnormal accounts and missing vouchers, then drafts a risk report, cutting the task from three hours to about one minute, the company said. POSCO DX projected that expanding AI agents to the entire financial close process would raise productivity by at least 30%.
Choi Tae-hwan, head of management support at POSCO DX, said, "An AI-native company is one where everything is designed with AI as the premise," adding, "Leveraging our strength as the only full-stack company in Korea with understanding and capabilities in both factories (physical domain) and management (office domain), POSCO DX will rise to become the company that implements physical AI best."