← Back to Insights

How I Built 54 Live Experiences With AI-Assisted Execution

I didn't write a single line of code. Here's how I built every one of them anyway — the workflow, the discipline, and the judgment calls that AI cannot make for me.

I want to be precise about something before I start.

When I say I didn't write a single line of code, I mean it literally. Not "I barely wrote code." Not "I had help writing code." I cannot write JavaScript. I cannot write PHP. I'm a Product Owner who spent years writing requirements, not software.

And yet, I have shipped 54 live experiences here.

This article is about how that happened, exactly.

Where It Started

In 2024 I was running a sprint planning session for a fintech team. We needed to estimate a backlog of 14 stories. We were using physical cards, Fibonacci sequence, the whole ritual.

And I thought: there has to be a better way to do this for distributed teams.

So I described what I wanted to an AI — a digital planning poker tool where remote team members could simultaneously reveal their estimates. And 40 minutes later, I had a working prototype.

Not a wireframe. Not a mockup. A working HTML page I could open in a browser and use.

I stared at it for a while. Then I thought about every other problem I'd been solving manually for years.

The Workflow

Every tool followed the same five-step process:

What Each Tool Actually Took

People assume these tools took months. They didn't. Here's a representative sample:

90 hours across 6 weeks, alongside full-time delivery work. A traditional development engagement for a tool suite this size would have been measured in months and tens of thousands in development costs.

What AI Is Actually Good At

AI is excellent at:

• Translating structured logic into working interfaces
• Applying consistent visual style across many components
• Building the "plumbing" — form handling, calculations, state management
• Generating variations quickly for testing
• Implementing established patterns (Fibonacci sequence, RICE formula, Kano categories)

What it genuinely cannot do — at least not without significant human guidance:

• Know whether the output of a calculation feels right to someone who has run that calculation on 50 real projects
• Decide which edge cases matter in a delivery context
• Judge whether the user experience matches the stress conditions under which the tool will actually be used
• Know when a mathematically correct answer is practically useless

The 150 iterations across these tools were almost entirely about that second list. Not bugs. Domain judgment.

What Surprised Me

The biggest surprise was how directly my existing skills applied.

Writing a prompt to AI is structurally identical to writing a user story. You define the actor, the goal, the context, and the acceptance criteria. You include edge cases. You specify what "done" looks like.

The skills I'd spent years developing — precise problem definition, structured thinking, requirements elicitation, acceptance criteria writing — turned out to be exactly the skills that make AI prompts effective.

I needed to apply product and business analysis discipline in a new medium.

What This Changes

The traditional delivery workflow has a large gap between "I understand the problem" and "stakeholders can see a working version."

That gap used to be measured in weeks. Procurement, requirements review, development handoff, sprint planning, the build itself. By the time something existed that stakeholders could react to, the original problem had often shifted.

This workflow collapses that gap to hours. The prototype isn't production-ready. But it's enough to validate assumptions before committing engineering resources.

That's what these 54 live experiences actually demonstrate. Not that AI replaces developers — it absolutely doesn't — but that the feedback loop between problem and prototype has fundamentally changed.

The Technical Stack, If You're Curious

Everything here is intentionally simple: