Foundation stage · Module 2 of 15
Outcome for this lesson
Identify where AI can fail and write the human-review rules for your course project.
Bring forward: Bring the project card from Module 1.
Produce: A four-item human-review boundary covering facts, requirements, risks, and tests.
0 of 15 core modules complete
Progress is saved on this browser. Create a free membership to sync across devices.
Recommended prerequisite: complete Module 1, “What Is Vibe Coding? A Responsible Builder’s Starting Point,” and bring its artifact. You may still use this page as a reference.
Foundation stage
Build the next result
You now have a project, but a clear idea does not make generated output trustworthy. This module turns common AI failure modes into a written review boundary that stays with the project.
Your result for this module
A four-item human-review boundary covering facts, requirements, risks, and tests.
Continuous case study
Worked example: Review boundary for the tracker
The tracker must not invent a date format, silently drop an inquiry, expose contact details, or claim that a reminder was sent when it was not. Each claim needs a named check rather than a general instruction to be careful.
Do the work
Separate facts from generated suggestions
List facts that must be confirmed in primary documentation or the running system, including packages, APIs, settings, and platform limits.
Expose unstated requirements
Write accessibility, privacy, error, recovery, and business rules the AI cannot safely infer.
Name the critical journeys
Choose ordinary, edge, failure, keyboard, and recovery paths that must be tested.
Assign the human decision
State who reviews evidence and who can approve release. AI output can assist; it cannot accept accountability.
Copy-and-complete template
Create your module artifact
Facts to verify: Requirements the AI must not guess: Important user journeys: Risks requiring qualified review: Human release owner:
Keep sensitive information, passwords, API keys, payment data, and private customer details out of course notes and AI prompts.
Quality gate
Check the result before the quiz
- Every important claim has a verification source.
- Accessibility and privacy requirements are explicit.
- At least one failure and recovery journey is named.
- A human release owner is identified.
How this moves the project forward
Module 3 uses the review boundary to research the problem without turning AI summaries into unsupported facts.
Evidence checkpoint and lesson summary
Confirm the artifact, then answer the key questions
The questions cover a core idea, a realistic decision, and evidence from the lesson. Incorrect answers identify concepts to review; they are not a complete measure of your experience or ability.
Confirm the artifact and answer all three questions correctly to unlock the recommended next step.
Continue to Module 3: Research with AI Agents: Find, Compare, and Cite Evidence