liacs/agentic-llm — fall 2026

Agentic Large Language Models

LLMs that reason, act, and interact. Taught at LIACS, Leiden University, around a harness loop you'll build four times over.

View the syllabus →
REASON chain-of-thought: symptoms → ACT call tool: patient_history() OBSERVE 3 prior diagnoses found REASON reflect: matches case #2 ACT route to specialist agent

Agentic LLM = LLM + Harness. The harness is the code wrapped around a model: it manages memory and state, calls tools on the model’s behalf, and lets it observe the effects of its own actions. The course, and its textbook Agentic Large Language Models, is organized around three pillars:

01 · reason

Reasoning

Chain-of-thought prompting, self-reflection, world models, and metalearning that let LLMs think through complex problems.

02 · act

Acting

Tool use, retrieval augmentation, and real-world assistants in domains such as medicine, finance, and scientific research.

03 · observe

Interacting

Multi-agent systems, role-based collaboration, and emergent social behavior in simulated societies of LLM agents.

Course Information

  • Institution: LIACS, Leiden University
  • Credits: 6 EC
  • Format: Lectures on Tuesdays and Fridays, plus four hands-on programming case studies (in Python)
  • Teaching team: Max van Duijn, Michiel van der Meer, Aske Plaat, Niki van Stein — see About for bios

Check the Syllabus for prerequisites and an interactive breakdown of every part and chapter, and the Schedule for the week-by-week timeline.

Announcements

subscribe via RSS