The Year I Stopped Writing Code

It may sound shocking, horrifying, or funny, but I do not think I have truly written code in about a year. That does not mean I stopped engineering. Quite the opposite, actually. I still define the problem, make decisions, and remain accountable for the results. AI implements, but I remain in charge.

When I started developing software, the normal process for me was to search Google, find an example that was close, and adapt it. Ctrl+C, Ctrl+V, and several hours of debugging counted as writing code manually. You could write a decent portion yourself, but no one has infinite memory. Sometimes you spent days debugging a problem and when you finally solved it, you felt like a code god. Sometimes you solved a crazy logical puzzle and the business changed its mind 5 minutes later. Good times.

Early AI-assisted coding felt like fancy autocomplete. You could write a comment and it would generate a method from that, but anything complicated required plenty of correction. Over time, AI became the new Google. No more searching for 30 minutes just to find code that kind of does what you want. It still hallucinated and had limited conversational memory, but some iteration could get you there.

Agentic development is where it all changed for me. Instead of producing one response, an agent can inspect the codebase, use tools, make a change, run tests, discover it was wrong, and try again. It didn't completely eliminate mistakes, but it can now gather evidence and verify its work instead of leaving every step to me.

That was the tipping point. I was no longer copying AI-generated answers into the codebase; the agent could put the change where it belonged and test if it worked. It's still a skill to keep requests focused and review the outcome, but it dramatically tightened the feedback loop.

I eventually hit the point where I could say: "Wow, AI is a way better coder than I am in almost every way. Even if I could have done it this well, it would have taken me forever."

Since then, my projects have gained test coverage, performance improvements, and other refinements that time constraints once pushed aside. The dashboard looks good on mobile now?? We have end-to-end Playwright tests?? I can work effectively in languages and frameworks I rarely or never use because the agent handles all of the syntax while I concentrate on the problem and the expected behavior.

I also don’t need to know what every character does to take responsibility for the result. I have the agent test and review its work, I test the software myself, and I give the code a look. When something warrants a closer look, it gets one.

But working code does not need to match exactly how I would have written it. There are a hundred ways to solve the same problem. At some point, the implementation details stop mattering, and whether the software actually works matters more.

Agentic AI has fundamentally changed how I view software development. The important skill is no longer remembering how to write every line or implement it yourself; it is directing AI effectively: providing context, defining the problem, breaking work into manageable pieces, recognizing bad results, and verifying that the finished product actually works.

I occasionally miss writing code by hand and untangling the logical puzzles myself. But engineering was never about producing code. It has always been applying technical and domain knowledge to solve problems. End users do not care who typed the code; they care whether the software is dependable and solves their problem.

So, with zero shame, I can say that I rarely write code anymore. I have become addicted to solving problems at a pace that previously seemed impossible. Some days I genuinely feel like the meme-level 10x engineer. How did I just get that done in a week when that would have taken me 6 months a year ago?

You do not have to begin by trusting AI with everything. Start small, give it a well-defined problem, and make it prove its work.