The Biggest Mistake I Made Using AI to Code

I know, I know. In the last post I promised that 'next time' will be the time I share all of the practical optimization I use while coding with AI.

I almost wrote that article. But here’s the problem:

• My “optimal” workflow changes weekly.
• What works today may be outdated tomorrow.
• Listing practices without context risks sounding like random tricks.

This space is evolving at rocket speed. Practices that are "best" today might be ancient by the light of dawn tomorrow.

What I call principles are not that fragile. You can apply no matter what tools you use, now or in the future.

Today's post will be about one of those principles; a mistake that was hidden in my workflow, and took me a long long time (so longer than I want to admit) to realize.

I thought my coding background would give me an edge with AI.

Turns out, it did the opposite.

⸻

The Tech Nerd Trap

Gosh, I've forgotten how many times I started a project by talking with an AI "I want to build an app with XX backend and YY frontend". I never thought of it as an issue, because to me, that's like the default prompt I will use to start a project.

That felt natural. I’ve spent years choosing tech stacks, debating frameworks, and optimizing for speed, memory, or whatever the trend of the year was. Why wouldn’t I start there?

It turned out, that "obvious" choice quietly sabotaged me.

⸻

The Butterfly Effect

Developers (myself included) default to what we know.

We pick stacks based on familiarity, benchmarks, or shiny-new-tool excitement.

But when you’re working with AI, the real question isn’t “What am I comfortable with?” It’s “What is AI good at?”

LLMs don’t magically become experts in every language or framework. They reflect what’s most available online:

• Strong docs
• Large communities
• Clear patterns
• Popular, “boring” tech

If you force AI into a niche stack — too new, too old, poorly documented, or just unpopular with solo builders — you’ve just handicapped your most powerful teammate. And suddenly you become the one carrying the project.

The “optimal” stack in your head can become the downfall of your project in practice.

⸻

When Nerdiness Backfires

Here’s what happens:

• You save milliseconds on requests with your “superior” tech choice.
• But AI stumbles on missing docs or rare edge cases.
• Bugs pile up. You spend nights patching what AI can’t handle.
• Momentum dies.

And the irony?

People with no coding background — the so-called “vibe coders” — often avoid this trap. They accept the AI’s recommended stack. It’s mainstream, boring, but robust. And they ship.

⸻

A Better Way Forward

When you’re solo-building with AI, your stack isn’t just about you.

It’s about both of you.

My current practice:

• Research what the majority uses.
• Stick with widely adopted frameworks and tools.
• If in doubt, ask AI to recommend a stack for your use case — then sanity-check it against docs, community size, and clarity.

But that stack has some serious issues, and I really hate it.

That's, my friend, is what I called "being a tech nerd".

You don’t have to love the answer. You just have to be pragmatic.

The goal isn’t technical purity. It’s getting your MVP out the door. If it flies, you can always refactor or hire a team later.

Being a tech nerd will kill your idea before it has the chance to shine. I say that with love — because I was one too.

If your goal is to ship, the boring choice is often the smartest one.