Abstract: With the collaborative advantages of high-order cognitive reasoning of large language model （LLM） and autonomous execution of industrial agents taken into consideration， a collaborative framework between vertical LLM （V-LLM） for blast furnace ironmaking and industrial agents was established， and a nested information feedback and iterative updating mechanism covering millisecond， hourly， daily and monthly scales was designed to provide theoretical basis for the new intelligent paradigm of blast furnace ironmaking based on LLM and industrial agents. A universal implementation path of "one center and five levels" and a three-dimensional evaluation system involving V-LLM， industrial agent and their collaborative efficiency were put forward according to actual production scenarios of blast furnace ironmaking. Case verification was performed by developing the intelligent agents for blast furnace temperature monitoring， early warning and decision making， and existing major challenges and corresponding coping strategies were analyzed. The research results facilitated the development of intelligent blast furnace ironmaking into a full-chain closed-loop collaborative system featuring cognition， decision-making， execution， feedback and iteration， and technical references were provided for the large-scale application of LLM and industrial agents.