Agents and tool use, from the loop to MCP
What makes a system an agent, how tool calling works underneath, and the planning, memory and multi-agent patterns interviewers ask about. It covers MCP, the failure modes that sink agents in production (loops, bad tool calls, runaway cost), the guardrails that live outside the model, and how to evaluate an agent you can't run the same way twice.
What you'll be able to do
- Decide whether a task needs an agent or a fixed workflow, and sketch the agent loop with its stop conditions
- Explain tool calling end to end, and design tools a model picks and calls correctly
- Compare ReAct, plan-then-execute and multi-agent designs, and say when each one hurts
- Use the compounding-error arithmetic to argue for recovery, guardrails and permission tiers
- Evaluate an agent with outcome checks, trajectory checks and pass^k
Before this: prompts-and-context
Log in to read the rest of this lesson
What makes a system an agent, how tool calling works underneath, and the planning, memory and multi-agent patterns interviewers ask about. It covers MCP, the failure modes that sink agents in production (loops, bad tool calls, runaway cost), the guardrails that live outside the model, and how to evaluate an agent you can't run the same way twice. This walkthrough runs about 18 minutes, with runnable code you can edit and re-run as you read. A new account gets 7 days of Pro free, no card required.
Log in to continueThe first module of every track is free to read on the web: see what's open in AI interviews.