An agent is not one thing. It’s a model, a set of tools it can call, a way to decide what to do next, a memory of what has happened so far, and the checks that stop it from going off the rails. Take any one of those away, and you have a chatbot with ambitions.

In this workshop, we’ll build an agent from those pieces using TypeScript, Google’s Agent Development Kit, and Gemini. We’ll add each part step by step: typed tools the model can use, instructions that shape how it reasons, state that carries across steps, routing that hands work off to a specialist agent, parallel steps that combine their results, and a human checkpoint for decisions you’d rather approve yourself.

By the end, you’ll understand what each piece does, how they work together, and why the agent falls over without them.