Part 1 of 10: When AI Stopped Being Theory and Started Being Real

I’ve taught software development to over 600 students a year at Columbus State. I’ve coached engineering teams at fortune 500 companies and governmental agencies. I thought I understood where technology was heading.

Then a colleague showed me what he’d built in a single week.

It was better than a year’s worth of work by a traditional team. Not “pretty good for AI-assisted.” Actually better. More features. Fewer bugs. Better architecture.

I stared at his screen feeling something I hadn’t felt in my decade as an Agile coach: professionally obsolete.

The Machine Learning Professor Who Missed the Revolution

Here’s the embarrassing part: I wasn’t new to AI.

I’d written machine learning algorithms for my master’s degree. I’d watched the LLM explosion in March 2022 with professional interest. I’d even given new engineers access to GitHub Copilot at a previous client.

But I’d treated AI like a fancy autocomplete tool. Useful for boilerplate. Dangerous for architecture. Something to be managed and contained.

I was watching the revolution from the sidelines, taking notes for a lecture I’d give next semester.

My colleague wasn’t taking notes. He was building.

Five Minutes That Changed Everything

“Just try it,” he said. “Tell Claude to make a simple web app.”

This was August 25, 2025. I’d just started a Claude Code subscription that morning, mostly because I was tired of hearing about it from this colleague.

I typed something vague. Something like “build a task manager.”

Five minutes later, it was done.

Not a tutorial. Not a code snippet I’d have to integrate. A working application with proper structure, error handling, and styling.

I sat there refreshing the browser like an idiot, waiting for it to break.

It didn’t break.

The Lunch Hour That Should Have Been a Warning

That week, I started bringing Claude to lunch.

Not literally—though that would’ve been less weird than what I actually did. I spent my lunch breaks at my desk, treating Claude like the world’s most patient junior developer.

“Write user stories for a task management app.”

Done.

“Decompose those stories into smaller pieces.”

Done.

“Implement story 1.1.”

Done.

It was intoxicating. Every Agile coach has fantasized about a team member who never pushes back, never misunderstands requirements, and works at superhuman speed.

But here’s what I missed in my excitement: Claude was also driving projects into the ditch at superhuman speed.

I just couldn’t see it yet. I was too busy being amazed that the car was moving at all.

What I Wish Someone Had Told Me

If you’re a technical leader reading this and thinking about bringing agentic AI into your organization, here’s what I learned in those first naive weeks:

Lesson 1: Speed without direction is just expensive chaos.

Claude could write code faster than any developer I’d ever worked with. But “faster” only matters if you’re heading toward something real. I was having Claude implement user stories I’d made up during lunch. No customer research. No clear problem. Just the intoxicating feeling of watching code appear on screen.

Lesson 2: Your team’s existing dysfunction will be amplified, not solved.

I brought years of Agile coaching experience to this. I knew about incremental delivery, vertical slicing, and definition of done. But I threw all of it out the window because the AI made it feel like I didn’t need process anymore.

If your team struggles with requirements clarity or technical debt now, AI will make those problems worse, not better. Just faster.

Lesson 3: The learning curve is steeper than it looks.

Five minutes to a working web app sounds easy. But that’s like saying “walking is easy” because you watched someone take five steps. I was about to spend two months learning to walk—falling repeatedly, questioning my competence, and rebuilding my entire mental model of software development.

The Credibility Problem

I need to be transparent about something: I’m using Claude to help me write these articles.

Not because I can’t write—I’ve designed curriculum, written training materials, and taught technical concepts for years. But because I want to practice what I’m preaching. If I’m going to tell you about working with agentic AI, I should be working with agentic AI.

Also, it’s faster. And I’m learning things about prompt engineering and collaboration that I couldn’t learn by writing alone.

Every technical insight, every failure, every lesson in this series is mine. The narrative structure and polish? That’s collaborative. Think of it like having a really good editor who never sleeps and doesn’t mind when I ramble.

What Comes Next

I’m ten weeks into this journey now. I’ve gone from “five minutes to a web app!” excitement to “nothing works and Claude keeps lying to me” despair to something more sustainable and honest.

I’ve built development metrics. I’ve created orchestration systems. I’ve learned which problems AI is genuinely good at and which ones are expensive traps.

But I’m getting ahead of myself.

This is the story of how a software developer turned professor with a decade of Agile coaching experience tried to build software with agentic AI. How I failed repeatedly. How I eventually found something that works.

And why I think hiring managers and coaches need to understand this journey—because your teams are about to go through it whether you’re ready or not.

Next time, I’ll tell you about the time everything felt possible. The time I thought I’d revolutionized software development at my desk during lunch breaks.

Spoiler: I hadn’t.

But I learned something more valuable than revolution: humility.

---

This is part 1 of a 10-part series documenting my journey from Agile coach to agentic AI developer building the app pomofy.net. Part 2 explores what happened when I tried to actually build something real.

About the Author: I’m an Agile coach and software development professor with 10+ years coaching teams at major enterprises and teaching 600+ students annually. I’m documenting this journey in real-time, failures and all, because I believe technical leaders need honest perspectives on what agentic AI actually means for their teams.