How to Design an App with ChatGPT in 90 Minutes

Intro

After getting my toes wet, I finally dove in and tried to create an app with just a prompt. I tried all of the usuals (Space Invaders, Frogger, Teach Me Turing Machines, and more). In a couple of hours of playing I had a 50% success rate, with half of that 50% having an error on the first attempt, but was able to fix it.

Then I realized that we still have to do some requirements and design, so I asked Zai, my ChatGPT, how this was supposed to work. Zai wrote us this How To.

I will not get to try it out until this weekend, but a friend asked me how to get started, so I am going to post now. I will post a follow up next week after I’ve given it a try. In the meantime, please share your experience.

Why This Matters

---

Too many MVPs collapse when it’s time to scale because they were hacked together without stable contracts, clear boundaries, or a plan for evolution. By using a structured 90-minute design brief with ChatGPT, you can produce an MVP that works today and survives into v1.

---

The 90-Minute Design Brief Process

1. Purpose & Context (10 min)

• Define who it’s for, the promise, what’s out of scope.
• State your MVP success signal: a measurable, observable outcome.

2. Domain Nouns & Verbs (15 min)

• List your key entities (nouns) and actions (verbs).
• Give each a one-line definition for clarity.

3. Contracts Between Steps (25 min)

• For each action, define the CLI/API signature (inputs/outputs).
• Write JSON schemas for your main data structures.
• Assign version numbers to contracts (e.g., RenderSpec@v1).

4. Non-Functionals (15 min)

• Set performance targets, cost ceilings, and platform constraints.
• Define error handling and idempotency rules.

5. Testing Oracle (15 min)

• Identify golden sample inputs and expected outputs.
• Define sanity checks and zero-rows validation tests.

6. Security & Compliance (5 min)

• Note any data egress rules, PII handling, and license constraints.

7. Evolvability (5 min)

• Identify where AI or new features will be added later.
• Flag human-in-the-loop steps to be automated in the future.

---

Working with ChatGPT

• Iterate in sections: Don’t dump the whole brief at once—have ChatGPT co-write each part.
• Ask for examples: Let the bot suggest step contracts, schema layouts, and validation rules.
• Stay in control: Treat ChatGPT like a senior pair programmer—you decide scope and priorities.
• Refine quickly: Use the bot to generate, then edit together for clarity and accuracy.

---

Intern Mini-Experiment

We’ll test this process by having each intern:

1. Pick from 3–4 small app ideas (some canvas-only, some IDE-based).
2. Fill out the 90-minute design brief with ChatGPT.
3. Build the MVP directly from the brief.
4. Deliver:

• The completed brief.
• The working MVP.
• A short reflection on what worked and what didn’t.

Scoring: Brief completeness, bot collaboration evidence, MVP functionality, self-reflection quality.

---

Takeaway

A 90-minute design session with ChatGPT isn’t about building the whole app—it’s about producing a skeleton with strong joints. Contracts, naming conventions, and stable boundaries mean your MVP can evolve instead of being rewritten.

With practice, you’ll spend less time untangling ad-hoc code and more time shipping features that last.