Claude Code Built It in Two Days. Then I Looked More Carefully.

Three months ago I had never used an AI coding assistant on a real project.

I’d played around. Asked ChatGPT a few questions. Used Copilot to autocomplete a line here and there. But actually building something with AI as a core part of the development process? Zero experience.

And I had never touched an AI API. No LLM integrations, no prompt engineering, nothing.

So I decided to do both at the same time.

What I built

DevTeam AI Assistant is a .NET 8 console app that uses the Anthropic Claude API to analyze sprint retrospectives, assist with code reviews, and prioritize technical debt. I built it using Claude Code as my development assistant — learning how to work with an AI coding tool and how to integrate an AI API simultaneously, from scratch.

Claude Code got me to a working build in 2 days

Two days. No prior AI coding experience. From blank project to functional application, with unit test coverage. That kind of velocity stopped me in my tracks — because I know exactly how long that would have taken me alone.

Here’s what stood out:

The learning curve was far gentler than I expected. Claude Code met me where I was — I didn’t need to already understand prompt engineering to get started. I learned it by doing it.

Velocity on the initial build was genuinely faster. Boilerplate, scaffolding, repetitive patterns — handled. Hours of mechanical work compressed into minutes.

Debugging was sharper. Describing a problem out loud — even to an AI — forces you to articulate it clearly. Half the time the act of explaining it pointed me toward the answer.

It’s a strong thought partner for architecture decisions. Not because it always gets it right, but because pushing back on its suggestions clarified my own thinking.

Honestly? My first impression was that this changes everything.

Then I looked more carefully at what it had actually built.

More on that next week.