When Did You Last Write Code?

When was the last time you wrote code by hand?

For me, probably January or February of this year. I’ve made small edits since then, but sitting down and writing a feature myself? It’s been a while.

When the Codex app arrived in February, I fully adopted it. It became how I build. I describe what I want, give it context, review the changes, and check that the result works.

That happened faster than I expected.

For UI work, PxlShot screenshots and spoken prompts help me explain the change. I’ve written more about my workflow with screenshots and voice.

We Had a Good Run

DHH’s Rails World 2026 talk put words to something I’ve been feeling. After decades of writing code by hand, he talks about moving into a new role as a “professional maker.” You can find the clip on X and the LinkedIn discussion.

“I have retired from being a professional programmer.”

What stuck with me was how he talked about the years spent coding. Enjoy what that was. We got to build things, solve hard problems, and disappear into a good coding session for hours.

I loved that too. Getting a feature working after being stuck was a great feeling. So was finding a simpler way to write something.

I’m glad I learned to build that way. That experience still helps me decide whether the AI is making a good change or creating a mess. But I don’t feel a need to keep typing every line to prove I’m a developer.

The Model Debate Keeps Moving

Claude, Codex, Gemini. Every new release seems to restart the argument about which one is best. One handles a task better, another gets an update, and everyone tries again.

“The key is: if one model gets stuck or gives mediocre outputs, try another.”

That makes more sense to me than picking a team and defending it forever.

Codex is my default. What matters to me is whether it can finish the work and whether I can trust the result after checking it.

“Good code still has a cost”

Generating code has become cheap, but producing good software still takes work. You still have to solve the right problem and show that it works.

It Can Work Through Its Own Mistakes

The powerful part is the whole loop. An agent can read the code, make a change, run the tests, inspect a failure, and try a fix. I can give it access to several repositories and have it follow an issue from the frontend to the API to a shared package.

That saves so much time. A bug rarely cares where one repository ends and another begins.

I keep thinking: this is the worst AI coding is ever going to be. That’s the bet I’m making, anyway. Progress will be uneven, and a new model can still make an old mistake.

The hope is that future models get better at finding and repairing the problems today’s models leave behind. We can already use agents to attempt that work, with tests and review to check the fix.

I’m excited about that. It also gives me a reason to keep the code understandable and the tests useful. I want the next agent to have something solid to work with.

Are You Still Writing Code by Hand?

If you are, I get it. But if you’ve ruled out AI without giving it a real try, I think you’re taking a risk.

Kodak comes to mind. A Kodak engineer built the first digital camera in 1975. The company sold digital cameras too, yet struggled with the shift away from film and filed for bankruptcy in 2012. Seeing a change coming doesn’t mean you’ve adapted to it.

That’s the part I think about with AI. I don’t want to be so attached to how I learned to build that I stop learning.

Knowing how to code still matters. So does being willing to change your workflow. Try an agent on a real feature. Review the changes. Run the tests. Find out what it can do for you.

I’d rather learn the tools now than spend the next few years defending how I used to work.

Ideas, Execution, and Marketing

Coding was a huge part of what I loved about building software. It still is, but whether I’m working on T.LY or WeatherExt, more of my time now goes into deciding what to build, getting it right, and helping people find it.

An idea is easier to try. That’s why I’ve called AI the best co-founder. Execution still means shipping something people can use. And marketing is still the hard part. A working app doesn’t come with customers attached.

Those things always mattered. Now I have fewer hours of typing standing between me and finding out whether an idea is any good.

I loved writing code. I like how much I can build now even more.