Exercise 10 of 10 · 0 passed
The Agent Loop
An agent is a model in a loop: think → act → observe → repeat.
The ReAct pattern (Reason + Act):
- Send task to model with available tools
- Model returns a final answer or requests a tool call
- Execute the tool, send the result back
- Repeat until done
This is the foundation of every AI agent — from arithmetic helpers to multi-step research pipelines.
Why care? This loop scales to any number of tools. Replace add and multiply with search_web, query_database, or send_email — the loop is identical. Understanding it here means you understand every agent framework under the hood.
Task
Complete Agent.run(task). Two arithmetic tools (add, multiply) and their schemas are already defined in TOOLS and TOOL_FUNCTIONS. Loop until the model returns stop_reason == "end_turn".
Click "Run tests". First time loads Python in your browser (~5s). After that, runs are instant on this page and ~1s on the others.
Loading tests…Pass the tests to reveal the reference solution.
The point of the exercise is the struggle, not the answer. Once your tests are green, the reference solution unlocks here so you can compare patterns.
Loading solution…That's the warm-up.
You ran ten exercises against a mocked SDK. The cohort builds the real thing: an AI agent shipped to production with 150+ passing tests, Telegram bot, FastAPI, Streamlit, Docker. Six weeks. Every PR reviewed.