Tracing

Agents chain many steps; without tracing you cannot see which step failed or slowed down. Essential for production debugging and interview system design.

Tracing is a package tracking number that shows every warehouse and truck stop — not just 'delivered' or 'lost.'

Scenario

User asks for refund status. Trace shows retrieval returned wrong policy doc — root cause found in 2 minutes instead of hours.

Solution

In Production Agent Engineering, apply Tracing to this scenario: User asks for refund status. Identify the inputs, run the technique, validate the output, and note one thing you would monitor in production.

Do this before moving to the next module — reading alone is not enough.

Open the Code Walkthrough below and run it locally. Change one parameter related to Tracing (e.g. model, temperature, top_k, or tool name), observe the difference in output, and write 2–3 sentences explaining what changed.

Highlighted lines show where Tracing happens in the code.

1# Tracing — minimal example2from openai import OpenAI3
4client = OpenAI()  # create API client5
6# Ask the model to explain this topic7response = client.chat.completions.create(  # core API call for Tracing8    model="gpt-4o-mini",9    messages=[10        {"role": "system", "content": "You explain tracing clearly."},11        {"role": "user", "content": f"What is tracing?"},12    ],13    temperature=0,14)15print(response.choices[0].message.content)  # show output for debugging

• Treating Tracing as a black box without evaluation
• Ignoring cost and latency in production
• Skipping error handling for tracing

Quick recap — the most important points from this module.