How I Used AI to Go from Zero → Product Concept → Prototype → MVP

Written By Gabriella Miele

I’ve been thinking a lot about how product managers actually use AI, not just as a gimmick, but as a real extension of how we think, design, and build.

I’ve done a lot of this type of work at LeafLink, but to protect the company, I decided to revisit an idea we had previously dropped. At the time, we couldn’t validate it quickly or get solid feedback without going through the full product lifecycle—discovery, hypothesis, and testing—which would have been too resource-intensive before seeing the potential value.

Instead of treating AI like a tool for answers, I treated it like a collaborator. What came out of that was a full end-to-end product concept: from raw problem framing, to structured requirements, to a designed prototype, all the way to a scrappy MVP.

Here’s exactly how I did it.

The Beginning: Turning a Messy Problem into Something Structured

I started with a problem I know very well: catalog management.

Specifically, the gap between how products are managed in a B2B system and how they actually show up for consumers. It’s messy, inconsistent, and honestly—kind of painful to scale. I also wanted to add the complexity of how this would translate in a B2C lens because oftentimes, the customers at LeafLink from the buyer side also had the same questions as our sellers: how do I get people to buy my product?

Instead of jumping straight into solutions, I used ChatGPT to help me think. First, I gave it all my perspective from the customer problem, to the type of impact we want to make in general, to the pain points of why this matters for us. Then, I told it more about the workflows that I saw as high impact, etc. Then I told it about the success metrics and the core KPIs we would want to work with first that would generate value to the company. Then, I asked it to question me, give me feedback, and validate everything that we just spoke about.

Not in a “give me the answer” way—but in a pressure-testing my thinking way.

I prompted it with:

  • What I wanted was to help me solve and ask what still feels unknown that I need to answer?

  • What are the core differences between B2B and B2C product experiences? Where does product data break down?

  • What would a “trusted catalog” actually need to include?

And then I kept going. Iterating. Challenging. Refining.

👉 Here’s the actual thread of how I worked through it:
https://chatgpt.com/share/69b9da70-c3e4-800c-8f21-8c169853bf56

What AI helped me do here wasn’t just ideation—it helped me:

  • Structure ambiguous thoughts into clear problem spaces

  • Identify gaps I wasn’t explicitly naming

  • Push past surface-level solutions

By the end of this phase, I had something much more valuable than ideas: I had clarity.

One thing I didn’t account for and assumed throughout this exercise is that I should scope this project down. It would be a ton of work for an engineering team, but because I am experimenting with Claude Code here as much as FigmaMake, I wanted to see what limits I could push it to.

The Middle: Translating Thinking into Real Product Requirements

Ideas are easy. Turning them into something a designer (or engineer) can actually use? Way harder.

So I used ChatGPT as a bridge.

I literally prompted it the way I would brief a designer:

  • “Write this as requirements for a bulk catalog management tool”

  • “Translate this into a consumer-facing experience”

  • “What states, edge cases, and flows am I missing?”

What came out of that process was surprisingly solid:

  • Clear feature definitions

  • Thoughtful user flows

  • Consideration of edge cases (which is where most specs break)

  • A foundation that could actually be designed against

Then I took that output and moved into Figma. I actually asked ChatGPT to guide me on the prompts to use that would resonate with a designer and an AI.

👉 Here’s the prototype I built off of those AI-assisted requirements:
https://www.figma.com/make/wtt9HGEpEe1Prhp0wmefvm/Bulk-Catalog-Manager-Design?t=dxvCCIqjITdzouIs-1

What mattered here is that AI didn’t replace product thinking—it extended it.

Instead of staring at a blank page, I was iterating on something structured. Faster cycles. Better questions. Less guesswork.

If thus was an actual project, this is where I’d start doing customer testing and validating, plus pressure testing with GTM on if we’re close or not.

The End: Pushing It Into an MVP

Here’s where it got really interesting.

I took everything—problem framing, requirements, flows—and used Claude to generate what an MVP version of this product could actually look like end-to-end. I have a connector between Figma to Claude to push it and not copy and paste it.

Not production-ready, obviously. But tangible.

Something you could:

  • Click through

  • Evaluate

  • React to

And that’s the key shift.

AI let me go from:

  • Abstract idea →

  • Structured thinking →

  • Designed experience →

  • Functional concept

…without needing a full team at every step.

👉 Here’s the output from Claude:https://www.loom.com/share/49878452202e4626947242cbaa0dcbb5

What I Actually Learned

This whole exercise changed how I think about AI in product work.

A few things that stood out:

1. AI is best as a thinking partner, not a shortcut
The value wasn’t in the answers—it was in the back-and-forth. The iteration.

2. It compresses the “blank page” problem
Going from nothing → something is the hardest part. AI makes that step dramatically faster.

3. It raises the floor, not the ceiling
You still need product intuition. Taste. Judgment.
But it helps you get to a solid baseline much quicker.

4. It connects the dots across disciplines
Product → design → (pseudo) engineering became one continuous flow instead of siloed steps.

Where I’m Going Next

This wasn’t just a one-off experiment.

I’ve been using AI products for a while now. But something I look forward to doing more of is using the prototypes I build to build real products and get them out faster, saving my engineers' time (and the company I work for) money.

It’s about thinking better, earlier, and more collaboratively—even when you’re working solo.

——

Here’s the prompts and how I used ChatGPT to help me build requirements: https://chatgpt.com/share/69b9da70-c3e4-800c-8f21-8c169853bf56

Here’s my figma file that i would prototype: https://www.figma.com/make/wtt9HGEpEe1Prhp0wmefvm/Bulk-Catalog-Manager-Design?t=dxvCCIqjITdzouIs-1

Here’s the actual build out from Claude if it were a new product MVP for the end-to-end experience: https://www.loom.com/share/49878452202e4626947242cbaa0dcbb5

Gabriella Miele
Next

Resume