Workshop · Agentic AI
Agentic AI workshop: build an AI agent, then a team of them.
No frameworks and no magic. Students write the agent loop themselves, give it real tools, put guardrails around it, then wire several agents into teams: a pipeline, a writer and critic, a router, a manager with workers, and a small software team that plans, codes and tests.
Book this workshop- Length
- About 3 hours: one long session, or two of 90 minutes
- For
- Engineering and computer science students, 2nd year and up; Department workshops, tech fests and AI clubs; Faculty development programmes on generative AI
- Students need
- Basic Python: functions, lists, loops and dictionaries; A laptop with Python 3.9 or newer
- Setup
- Runs with an OpenAI API key, or completely free on each laptop with Ollama and an open model, so no student needs a paid account.
We build the course the workshop runs on
A workshop is only as good as what students have in front of them. For this one we made three things:
Visual lessons for the projector
Animated lessons that show what happens inside an LLM call, the messages list, tool calls and the agent loop, stepping through the code line by line before students type it.
Open the visual lessons ↗A code-along repository
Small, numbered Python files that each add one idea, two labs with TODOs for students to finish, and worked solutions.
See the repository on GitHub ↗A facilitator guide
A minute-by-minute run of the session: what to run, the one line to say at each step, mistakes to trigger on purpose, and what to do when a model misbehaves live.
What students build, step by step
Module 1 · Build one agent
From a single model call to an agent that searches, calculates, writes files and runs code.
- 01Talk to the model, and see what it can’t do on its own
- 02Memory is just a list of messages
- 03The first tool: the model asks, your code runs it
- 04The agent loop: think → act → observe, until done
- 05Real tools with guardrails: a sandbox, human approval, step limits
- 06Package it as a reusable Agent class
- 07Lab: build your own to-do agent
Module 2 · Build a team of agents
The four shapes every multi-agent system is made of, and when to use each.
- 01Pipeline: researcher → writer → editor
- 02Reflection: a writer and a critic loop until the work passes
- 03Router: triage sends each question to the right specialist
- 04Orchestrator: a manager agent uses other agents as its tools
- 05Capstone: planner, coder and tester ship working, tested code
- 06Lab: build a trip-planning team that stays on budget
By the end, students can
- Explain what makes a program an agent, not a chatbot
- Write tool definitions and the agent loop without a framework
- Add guardrails: sandboxing, approval, limits and error recovery
- Choose between pipeline, reflection, router and orchestrator designs
- Recognise these same pieces in LangGraph, CrewAI and agent SDKs
Run it at your college
Tell us the date, roughly how many students, their year, and whether it’s in person or online. We can run it, or hand your faculty the materials and the guide to run it themselves.
Book this workshop