Build AI Agents
A model, a loop and some tools - and how to trust it with real work.
- 1A model, a loop and some toolsAn agent is a language model that can ask your code to do things, in a loop, until it has an answer. Step through one call at a time, watch the context grow, and find out who actually decides when it stops.Runs in your browserOn your machine17 min+35 XP
- 2Tools: how a model touches the worldA model picks tools by reading their descriptions, and anything a tool returns can talk back to it. Write a description that works, then stop an email from emptying the refunds account - without asking the model nicely.Runs in your browser18 min+45 XP
- 3Workflows first, agents when you mustMost jobs that sound like they need an agent are a fixed sequence of model calls with a few branches. Run 20 support tickets through a workflow, an agent and a mix of the two, and compare what each one handles, costs and keeps you waiting for.Runs in your browser14 min+35 XP
- 4Guardrails, approvals and evalsA demo shows an agent working once. An eval shows whether it still works after you change it. Improve a support agent's prompt, watch the demo stay green while two real cases break, and fix it with a line of code instead of a sentence.Runs in your browser18 min+40 XP
- 5Agents that help run a businessChasing invoices, triaging support, following up leads, sorting the inbox - an agent can take hours of this off a small team. Decide what it may do alone and what needs a person, then price it with real numbers.Runs in your browserOn your machine16 min+40 XP
Next track: SI Frontiers →