AI Made Me Faster at Coding. So Why Did I Feel Like a Worse Developer?

I’ve been working with React, Node.js BFFs, and Python for years. We use them at work, so there hasn’t been much reason to pick up something new.

But with AI making it easier to get started, I figured, why not try Svelte and Go?

I watched a few YouTube videos, learned the basics, and started building a dashboard for myself. Nothing too fancy. Just my Jira tickets, pull requests, code coverage, build status, and meetings in one place.

You know, instead of opening twelve tabs to find out how busy I am.

Everything worked. That was the problem.

I took it one feature at a time with Copilot. Once something was ready, I tested the UI and asked it to document the changes. If something broke, I shared the logs or explained what was happening, and it helped fix it.

A few days later, the dashboard was done. I was pretty happy with it.

Then I actually sat down and looked through the code.

And… I barely knew what I was looking at.

I understood the architecture. I had discussed those decisions with Copilot, and I knew how the app worked at a high level but following the actual Svelte and Go code was a different story.

If I had to fix a bug or add a small feature myself, I would have to debug the whole thing first. Which is funny, because learning how to do that was the reason I started the project.

I hadn’t really learned Svelte and Go. I had mostly learned how to ask Copilot to build things in Svelte and Go.

Small difference. Slightly important.

I missed figuring things out

Just to be clear, this was a side-project experiment, not how I handle feature development at work.

Still, it bothered me. I used to enjoy digging into a problem, getting stuck, and finally figuring it out. Sometimes it took hours. Sometimes days. But getting it working felt good because I understood why it worked.

This time, I had the finished project without much of that satisfaction.

Instead, I was already thinking about what else I could build. There was this weird FOMO around not doing enough with AI, even though I had just finished something.

I also wondered how much practice I was giving up by letting AI handle every difficult part.

So I started doing things differently.

Here is what's been helping

• Understand what I’m shipping. Speed is nice, but I don’t want a codebase I’m afraid to touch without opening a chat window.

• Break the feature into small tickets. I give AI the context and ask it to split the work into manageable pieces. Not “here is my idea, see you when the app is done.”

• Talk through one ticket before coding. I use plan mode and ask it to ask me clarifying questions. I question suggestions that don’t make sense, and sometimes get another model to review the plan.

• Stop when something is unfamiliar. If it suggests a circuit breaker, for example, I go learn what that is and whether I actually need one.

• Implement, then review. I use a cheaper model when it makes sense, and sometimes ask the planning model to check the implementation.

• Read the code and test it myself. If I don’t understand something, I ask for an explanation or step through it in the debugger. This is much easier when the change is small.

• Commit and move to the next ticket.

It definitely takes longer than letting AI run with the whole thing. But it’s still quicker than my old process, and I’m not skipping the part I wanted to learn.

I’m still using AI because AI is here to stay. I just want to stay involved enough that the code feels like mine.

Because if it breaks in production, I’m the one who has to explain what happened. Copilot isn’t joining that meeting.