Phase 3 — Core AI Engineering · Lesson 40 · 40 XP
Project: tool-using agent with tracing
The second Phase 3 capstone: a tool-using agent (Lessons 33-34) that does something genuinely useful — research, a coding helper, or a data analyst — instrumented with tracing (Lesson 36), with a real failure-analysis write-up covering at least one run that went wrong and why.
A good failure write-up distinguishes where the failure actually occurred: bad reasoning by the model, a broken or misused tool, or a bug in your own orchestration code. Tracing is what makes that distinction possible instead of guesswork.
Exercise
Build a tool-using agent that does something useful, with tracing instrumented per Lesson 36, and a short failure-analysis write-up covering at least one case where it went wrong — the second Phase 3 portfolio project.
Check yourself
1. In your failure write-up, was the failure in the model's reasoning, a tool, or your orchestration code? How did tracing help you tell the difference?
2. What's one guardrail (Lesson 37) this specific agent should have, given what it's allowed to do?
Project: RAG app with an eval suite
Answer the check-yourself questions to unlock this