Lesson 2 of 4 · Free to read
Exercises
Exercises — Visual AI Builders: Idea to MVP
Watch & learn
Course video
Course video
Keep a prompt log while you work. Speed without a trail is not a skill — it is luck.
How to use this file
- Each exercise is time-boxed and ends with a measurable “done” signal.
- Write your prompts and decisions in a short evidence log (a simple markdown file is enough).
- Avoid paid services. Use free tiers or local stubs. If a tool asks for billing, pick the stub path.
Exercise 1 — MVP spec + acceptance criteria (before you generate)
Goal: Produce a mini spec that can drive a builder prompt without scope creep. Time: 20 minutes Level: foundational Modules touched: [19-visual-ai-mvp]
Setup
- Pick one of these MVPs: “campus events board”, “micro habit tracker”, “study group matcher”, or “office hours scheduler”.
- Create a file
EVIDENCE.mdin a local folder (or shared doc).
Steps
- Write a 1–2 sentence problem statement.
- Write exactly 3 acceptance criteria (observable, testable).
- Write 3 “out of scope” items.
- Write 2 constraints (e.g. “don’t change existing routes”, “no paid services”).
Success criteria
- Your
EVIDENCE.mdhas problem statement + 3 acceptance criteria + 3 out-of-scope + 2 constraints.
AI prompt hints
- "Ask Cursor: 'Rewrite these acceptance criteria so they are testable and not vague. Keep scope small.'"
- "Ask Cursor: 'Suggest 3 negative tests (failure cases) for this MVP.'"
Solution sketch
Treat the spec like a test: if the builder output can’t be checked against criteria, the criteria are too vague. Constraints prevent UI churn and accidental rewrites.
Stretch
- Add one security boundary as an acceptance criterion (e.g. “anonymous users can’t edit”).
Exercise 2 — First build + “happy path” demo URL
Goal: Produce a working demo URL that satisfies the happy path. Time: 40 minutes Level: foundational Modules touched: [19-visual-ai-mvp]
Setup
- Open your visual AI builder of choice.
- Paste your spec and constraints from Exercise 1.
Steps
- Generate the first version.
- Run the “happy path” manually (as a user would).
- Publish the demo URL.
- Capture 2 screenshots: home page + a successful action.
Success criteria
- You have a stable demo URL and a recorded happy path (screenshots or a short screen recording).
AI prompt hints
- "Ask the builder: 'Generate the MVP using only free/local stubs; do not require billing setup.'"
- "Ask the builder: 'List the top 5 assumptions you made and how to validate each.'"
Solution sketch
The goal is working, not perfect. Publish early so you can iterate with a constant “deployment” target (a URL).
Stretch
- Add a lightweight “About / limitations” page that lists known gaps.
Exercise 3 — Data-flow diagram (UI → API/logic → storage → response)
Goal: Explain your MVP’s core logic without reading code aloud. Time: 25 minutes Level: intermediate Modules touched: [19-visual-ai-mvp, 04-software-design]
Setup
- Create a diagram in a doc or on paper. Four boxes is enough.
Steps
- Draw the 4-box flow.
- Label the storage (even if it’s fake/local).
- Add one “failure path” (e.g. validation error).
- Add one trust boundary (e.g. “anonymous vs authenticated”).
Success criteria
- A readable diagram exists and matches what you can demonstrate in the UI.
AI prompt hints
- "Ask Cursor: 'Given this MVP description, propose a 4-box data-flow diagram and one trust boundary.'"
- "Ask Cursor: 'Spot likely inconsistencies between this diagram and a typical builder-generated app.'"
Solution sketch
The diagram is the contract between what you claim and what the system actually does. It also exposes when you need to switch to code.
Stretch
- Add a second diagram: “future state” if you moved to a codebase.
Exercise 4 — Add a non-trivial feature with constraints and rollback
Goal: Add one feature while avoiding regressions and capturing decisions. Time: 45 minutes Level: intermediate Modules touched: [19-visual-ai-mvp]
Setup
- Pick a feature that affects more than one screen (e.g. “filter”, “role-based edit”, “export to CSV”, “reminder email stub”).
Steps
- Write 2 new acceptance criteria for the feature.
- Write 2 “must not break” criteria for existing behavior.
- Prompt the builder with constraints + criteria + rollback instruction.
- Re-run the happy path and the negative test from Exercise 1.
Success criteria
- The new feature works and the previous happy path still works; your evidence log includes the prompt and the validation steps.
AI prompt hints
- "Ask the builder: 'Implement feature X. Do not break existing behavior Y and Z. If you need to change existing routes, explain why and propose a rollback plan.'"
- "Ask Cursor: 'Rewrite this prompt to be more constrained and less likely to cause unrelated changes.'"
Solution sketch
The key is preventing accidental rewrites. “Must not break” criteria and rollback guidance make the tool safer even when it guesses.
Stretch
- Ask for a diff-style summary of changes (“files touched + why”) and add it to your evidence log.
Exercise 5 — Builder → code handoff plan (minimal, credible)
Goal: Decide what you would keep vs rewrite when moving to a real repo. Time: 30 minutes Level: intermediate Modules touched: [19-visual-ai-mvp, 02-ai-assisted-development]
Setup
- Export/download your project if the builder supports it.
Steps
- Identify: routes/pages, API handlers, DB schema/config, env vars, and any auth integration.
- Write a “handoff README” with: run steps, env vars, known issues, and the first 3 refactors.
- Decide what gets rewritten first (pick exactly one component or endpoint).
Success criteria
- A
HANDOFF.mdexists with run steps + env list + first 3 refactors + “rewrite first” choice.
AI prompt hints
- "Ask Cursor: 'Given this builder app, what are the minimum files/concerns to extract so a dev can rebuild it in code? Produce a checklist.'"
- "Ask Cursor: 'Propose the first refactor to do after export, and why it reduces risk.'"
Solution sketch
You want a handoff that a teammate can use without you. The first rewrite should reduce risk (auth boundary, persistence, validation) rather than “make UI nicer”.
Stretch
- Convert your evidence log into a short demo script (2 minutes) with “what works” and “what’s next”.