It Wasn't Claude. It Was Me.

For a week I was unstoppable.

I'd describe what I wanted, the code appeared, everything worked. The pace was absurd. The feedback loop so tight it almost didn't feel real.

Then it didn't.

The code got messy. The system got harder to reason about. I'd open files and think: what is going on here?

And right on cue, the excuse arrived:

"AI wrote this. Of course it's confusing."

Except that wasn't true.

The truth was way more annoying.

It wasn't Claude. It was me.

The problem wasn't the code. It was the target.

I kept changing requirements while the system was being built.

Not in a thoughtful product way. In a "while I'm here..." way.

Add one more thing. Then another. A small tweak that isn't small. A new edge case. A slightly different flow because I thought of a cooler version. A new setting because maybe this should be configurable.

None of these are bad ideas in isolation.

Together, they're a moving target. And moving targets don't create clean code. They create coping mechanisms.

The model did exactly what I asked. Over and over. In the order I asked it. With all my mid-flight changes included.

That's how spaghetti is made.

"The model didn't build spaghetti. It just cooked faster than I could taste."

AI didn't create a new failure mode. It surfaced an old one instantly.

In the old world, this same behavior took weeks to punish me.

Shipping was slow. The consequences of vague thinking and shifting requirements took time to accumulate. You could coast. You could promise yourself you'd clean it up later.

AI removes the delay.

When implementation gets cheap and fast, the system reflects your decisions immediately. What used to take weeks to surface shows up right away.

That's not the model being wrong.

That's the feedback loop being honest.

"AI didn't confuse me. I confused me."

"I don't understand the code" is a proxy for something else.

Sometimes we blame the tool because it's the most visible
thing in the room.

But "I don't understand the code" usually means:

I don't understand what I'm building anymore. I changed the goal three times and now the code represents all three. I asked for five features that don't belong together. I didn't choose a direction, I collected ideas.

AI didn't do that. I did.

The model just kept up.

The new bottleneck is your decision-making.

Engineering speed used to be the constraint. Code took time. Refactors were expensive.

Now the code can move faster than your product thinking.

So the bottleneck becomes clarity. Taste. Sequencing.
Restraint. Knowing what to cut. Knowing when to stop.

Unsexy. But it's the part that determines whether AI makes you feel unstoppable or makes you feel lost.

The good news: refactoring is no longer a tax.

Refactoring used to feel like punishment for past decisions. Something you only did when things were on fire or when you had that rare mythical week called "engineering time."

Now you can refactor at a rate that feels unfair.

Simplify aggressively. Restructure large pieces and still keep momentum. Delete whole approaches and rebuild them cleanly in the same day.

The new play: iterate fast, then simplify fast.

Two modes that have been working for me.

Mode 1: Make it exist.

This is exploration. Speed is the point.

Build the end-to-end flow. Accept hacks but label them. Prove the product shape. Learn what you actually meant.

Admit it's messy on purpose.

Mode 2: Make it make sense.

No new features. This is where the real engineering happens.

Delete the "while I'm here" features. Collapse duplicate logic. Name things like you want to live with them. Make the data model boring. Add a small amount of testing around the core path.

Most importantly: no new features during simplification.

If you keep inventing while you refactor, you're not refactoring. You're continuing the chaos with better formatting.

I know because I keep trying.

"If you want AI to make you a better builder, don't
obsess over every line it wrote. Obsess over something
harder: are you making good decisions?"

AI will happily execute your chaos at lightning speed.

And it will just as happily help you simplify it.

If you give it the chance.