🤖 Why Agile + AI Coding Feels Kinda Broken

So I’ve been thinking about this a lot lately (probably too much lol). Everyone in dev land keeps saying: “We work Agile. We do sprints. We build MVPs first. Nice-to-haves later.”

But the second you start coding with AI? That whole system feels… off.

Agile vs. AI Speedrun

Agile was made for a world where coding was slow, expensive, and full of risk. You break things into stories, you prioritize, you sprint, you iterate. It makes sense… if typing code is the hard part.

But with AI, typing isn’t the bottleneck anymore. Thinking is.

The AI doesn’t care what’s “MVP” or what’s “nice-to-have.” It’ll happily generate both in the same session. Sometimes I just ask for a basic version, and it still goes “oh btw, here’s a dashboard, an auth flow, and five utility functions you didn’t ask for.” Thanks? I guess? 😂

So instead of carefully adding things sprint by sprint, you kinda… teleport to the end state. And then throw half of it away because you realize you didn’t actually need all that. That’s why Agile feels weird with AI — the cycles are too fast.

Building From Scratch With AI

When I start a new project now, I don’t think like an Agile dev. I think like:

Can AI generate me an overall plan?
Can it break that plan into pieces?
Which parts do I actually care about, and which can I let the AI cook?

It’s not about backlog grooming anymore. It’s more like blueprinting with a co-pilot who over-delivers.

And honestly, that shift has been huge for me. It’s like pair programming with a super eager junior dev who codes 100x faster but sometimes invents random imports. You gotta guide it, babysit it a bit… but the speed is unreal.

Refactoring With AI

Another weird thing: refactoring doesn’t happen at the “later” stage anymore. With Agile, you ship now and clean later. With AI, I’ll generate a messy version and immediately ask it to refactor, add tests, or even port it into a different stack.

So instead of carrying “technical debt,” I carry “prompt debt.” The more I refine my instructions, the cleaner the code gets — without me manually rewriting thousands of lines. That part feels totally alien compared to how I used to work.

My Portfolio Story 🎨

When I built my portfolio site recently, I leaned fully into this workflow.

First, I had AI create an overall plan. Then I asked it to break that down into smaller specs. Boom — I had a roadmap without touching Jira.
Next, I got AI to help me build a design system: colors, components, typography, the whole vibe.
Then I converted that into a Next.js site.
Added my own content (the only boring manual part, lol).
And finally… just let the AI cook. 👨‍🍳✨

The result? A working site full of small bugs, lots of “any” types, messed up margins and colors not following the system design. Without me changing the code its not usable.

So What’s The Deal?

For me, Agile with AI feels like forcing an old rhythm on a new instrument. The methodology was built for scarcity — AI dev is built for abundance. You don’t sprint anymore, you speedrun.

That doesn’t mean Agile is dead — it’s still great for teams, communication, structure. But if you’re solo-building with AI, the playbook changes:

Don’t overthink specs. Just get AI to sketch the whole thing.
Don’t fear “nice-to-have.” You can always throw things away later.
Think less like a project manager, more like a DJ mixing tracks.

And most importantly: remember the point isn’t to follow Agile “perfectly.” It’s to ship, learn, and keep building. With AI, that loop just got a whole lot faster. 🚀