STRATEGIST: How LLM Agents Learn Strategy Through Self-Play and Tree Search | Jonathan Li (Caltech)

Jonathan Li is an AI researcher focused on language model reasoning, strategic planning, and multi-agent learning. His work explores how AI agents can discover and refine high-level strategies through self-play, evolutionary search, and abstraction, enabling stronger long-horizon decision making with only a small number of interactions. In this talk, Jonathan presents Strategist, a framework that teaches language model agents to learn reusable strategies instead of optimizing individual actions. By separating high-level strategy from low-level policy, Strategist enables agents to adapt more quickly across complex multi-agent environments such as negotiation, social deduction, and collaborative planning. Strategist combines language models with bi-level tree search, self-play, and evolutionary search to automatically generate, evaluate, and improve strategy libraries. Jonathan demonstrates how these learned strategies help agents reason about other participants, model their intentions, and make stronger long-horizon decisions while requiring only a handful of interactions. The talk also explores why high-level abstractions may be a more scalable path toward general-purpose AI agents than optimizing individual actions alone, bridging modern language models with classical search and decision-making techniques. Strategist paper: https://arxiv.org/abs/2408.10635 Topics: • Strategist • language model agents • multi-agent systems • strategic reasoning • long-horizon planning • self-play • evolutionary search • bi-level tree search • strategy learning • negotiation agents • social deduction • inference scaling Presented at Frontier Research Club by Jonathan Li. Recorded on June 10th, 2026, at Pebblebed. ​Pebblebed is a technical early stage VC founded by Pam Vagata (cofounder of OpenAI, ran AI for Stripe, inventor of FBLearner Flow); Keith Adams (founded Facebook AI Research, was chief architect at Slack, 20th engineer at VMWare) and Tammie Siew (former Sequoia Southeast Asia investor, former Sequoia & Notable Capital-backed founder). Frontier Research Club is a curated forum for rigorous discussion on how AI is reshaping the scientific research process. We convene researchers, computational scientists, and research engineers to examine concrete work across literature synthesis, hypothesis generation, experimental design, simulation, analysis, safety, and reproducibility. Upcoming events: https://luma.com/frontiersyndicate Subscribe for more research talks, technical discussions, and frontier AI presentations.