Exercises — Module 01: The agent loop

1.1 — Build the minimal loop

Implement the 10-line loop from Module 01 against a real LLM API. Add one tool: get_time() -> str returning the current ISO timestamp. No frameworks; raw SDK only.

Done when: you can run python loop.py "what time is it?" and get a sensible answer after one tool call.

1.2 — Break it, then fix it

Add a tool recurse_forever() whose handler always returns "please call recurse_forever() again". Run your harness. Observe.

Then:

• Add a max-turns guard.
• Add a wall-clock guard.
• Distinguish the stop reasons in your returned result.

Deliverable: the broken run log, the fixed code, and a one-paragraph reflection on how you would catch this in code review before a repeat.

1.3 — Error shaping

Add flaky_network() that fails with ConnectionError 30% of the time. Implement two variants:

• Swallow: the harness retries up to 3 times silently.
• Surface: the harness immediately returns the error as a tool result to the model.

Run 20 queries under each. Which gets better final outputs? Under what conditions would the answer flip?

Deliverable: numbers, plus a paragraph explaining them.

1.4 — Streaming and cancellation

Convert your loop to stream responses. Add a cancellation token checked between streamed chunks. Demonstrate Ctrl-C cleanly aborting a mid- generation run.

Deliverable: a short demo video or terminal transcript.

1.5 — Loop topologies

Read one paper for each of: ReAct, Plan-and-Execute, Reflexion, Tree-of-Thoughts. For each, write 3 sentences: what it is, when you’d use it, why you would not.

Deliverable: a 12-sentence note.