'Solving real problems with AI': Posco cuts R&D cycle from 32 weeks to 6

"In the past, we focused on what AI could do. Now we want to focus on what needs to be done — the goal is to innovate real work by concentrating on the core mission of our business."

Lim Chi-hyun, head of the group DX strategy division at Posco Holdings, outlined the group's AI transformation direction at the Herald Business Forum 2026, held Tuesday at Shilla Hotel in Jung-gu, Seoul.

Posco Group describes its approach as "mission-oriented AX" — a strategy that does not treat the adoption of AI technology as an end in itself, but instead focuses on solving real business problems and translating them into tangible results.

Posco Group has also built its own framework for converting AI-generated outcomes into measurable financial value. Rather than simply tracking improvements in model accuracy, the system asks how much in actual resources — raw materials, energy, time, risk — was reduced as a result, and what that reduction is worth in monetary terms.

"On the financial side, every company has its own formula for measuring performance," Lim said. "Think of it as mapping precisely how AI can contribute to that formula."

The framework, called AIReX, links AI-driven performance improvements to real changes in resource consumption, then factors in unit costs, scale of application and actual on-site usage to arrive at a monetary value — while also accounting for development, operating costs and labor input. The system attempts to quantify value not only for applications such as rolling-equipment fracture prediction and slab charging forecasts, but also for vision AI-based safety monitoring, where calculating financial impact had previously been difficult.

Lim identified research and development as one of the areas where AI transformation is producing results most quickly.

Posco Group has set a target of cutting its R&D cycle from the current 32 weeks down to six by applying specialized large language models, laboratory automation, virtual experiments and automated results analysis to the development of new steel and secondary battery material products and processes. Target areas include advanced automotive steel sheet products, optimal fuel usage methods, and solid electrolyte and cathode material process development for all-solid-state batteries.

Significant time savings are already emerging in some tasks. "We are seeing positive cases — a task that used to take two days now gets done in under three minutes, and the time spent analyzing experimental results has been cut by 95 percent," Lim said.

He added that the group is going further than simply deploying individual AI agents. "We are not stopping at building and using a single agent — we are trying to redesign the entire workflow from start to finish so that AI can carry out the work end-to-end, and we are even restructuring the organization to match," he said.

On the manufacturing floor, AI transformation is also expanding from automating individual processes to optimizing entire steelworks. "In the past we focused on point automation and point AX for each unit process — now we are moving toward AI overseeing the entire operation," Lim said.

Posco is scaling up an autonomous manufacturing model — where AI and robots collaborate with workers — centered on 11 "intelligent factory" model facilities at its Pohang and Gwangyang steelworks. The goal is what the company calls "seamless autonomous manufacturing," connecting every stage of production in a single unbroken chain from raw material unloading through ironmaking, steelmaking, hot rolling, heavy plate, cold rolling and plating, all the way to shipment.

In the blast furnace, the system has been refined to the point where AI incorporates the experience of skilled workers to assess conditions and predict abnormal situations. In steelmaking, the company is pursuing "one-touch operation," linking multiple unit processes through AI. "With one-touch operation, we have actually run fully automated operations from start to finish with no human involvement," Lim said. "Humans still collaborate on safety and more advanced tasks, but we are nearly at the level where AI can handle the work."

The company has also developed an AI agent that reads energy flows across an entire steelworks. Steelworks draw on a diverse mix of energy sources — external power, LNG generation and in-house generation using byproduct gases — and the optimal combination shifts constantly depending on fuel prices and equipment operating conditions. The AI agent analyzes production and power flows to help identify the lowest-cost option at any given moment.

"We have completed verification using historical data and plan to connect the system to the live production environment to optimize energy-related decisions and reduce costs on a daily basis," Lim said.

Posco Group plans to push ahead with 40 flagship AX initiatives across three areas — process, operations and value transformation — by 2028. The scope will extend across the group to include secondary battery process automation and remote operation, AI-assisted decision-making in trading and construction, and automation of human resources and financial operations.

"Korean manufacturers have accumulated decades of know-how," Lim said. "We want to turn that into an asset, encode it in AI, and use it to contribute to the country and the industry." The vision is to convert manufacturing experience built up over decades — in people and on the factory floor — into AI knowledge that will not disappear, and to translate that into new competitive strength.