Before you begin
Starting point: Experience describing workflows, evidence checks, and failure handling. Coding is optional for this design route.
Your output: An evaluation blueprint with explicit release and rollback criteria.
Read one guide, try its exercise, then explain the idea in your own words. Use fictional data while practicing.
Your reading sequence
- STEP 01
Bound the system
List tools, allowed actions, approval requirements, and stop conditions.
Read: AI agents & permissions → - STEP 02
Inspect the evidence boundary
Check source access, version selection, and unsupported-answer behavior.
Read: RAG explained → - STEP 03
Define the release gate
Separate quality dimensions and name failures that block a release.
Read: Evaluation & risk → - STEP 04
Build the evaluation record
Define cases, rubrics, critical gates, and the limits of your evidence.
Read: Build an AI evaluation → - STEP 05
Review the data journey
Map sensitive inputs, outputs, connected tools, and operational logs.
Read: AI data privacy →
Put the ideas together
- Prepare ten fictional evaluation cases with expected behavior.
- Compare two candidate designs using the same rubric and cases.
- Write a release recommendation, remaining uncertainties, monitoring owner, and rollback trigger.
Keep your first attempt and your revised version. Write a short note explaining what improved and what still needs checking.
Open the expert practice modules →The agent that passed the wrong test
Inspect an apparently strong evaluation, uncover a critical permission failure, and write a defensible release memo.
Work through the Expert case →Check your readiness
- I can explain why a critical failure outweighs an improved average score.
- I can show who owns permissions, monitoring, and incident response.
- I can justify a release decision using recorded evidence.
Use these statements as reflection prompts, not a formal score. If one feels difficult, repeat the relevant exercise with a different example.
Take the next step
After your core project, use these guides to connect the learning to an operational decision.
Choose your next step
Stay with this route until you can show your reasoning. You can switch levels whenever the work calls for it.
Beginner route → Practitioner route →