Over the past year I've had several moments where I thought: well, that's it, the work I used to do is gone.
I haven't written a line of code myself since December. When agents took over the writing, I was worried that only code review would be left. For a while that's more or less what happened. I read everything the agents produced, and guarding code quality and architecture became my main responsibility, because agents back then still made a mess of things if you let them.
That part is mostly gone too. With quality loops where separate review agents check the architecture, framework idioms and my own preferences, the code that reaches me is usually fine. I still look at it, but it takes a fraction of the time it used to.
What's left looks a lot more like the job of a product owner. Most of my day I'm dealing with questions like: does this application do what users actually need? Is it pleasant to use? Is this even a good product? I spend less time on how software should be written and more on deciding what good software is.
You'd expect that after losing most of my old job in less than a year, I would have a lot of free time. The opposite is true. I've never been this busy.
A hundred times more software
Before agents, I could build maybe one application in a certain period of time. Now I work on dozens of them in that same time. Some are work projects, some are websites for friends and family, some are little tools that only I use. But they all need someone to decide what they should do and whether they're any good.
Looking at it like that, you'd think we need far fewer programmers. If one developer can do the work of a hundred, the other ninety-nine can go home. But that only works if we're happy with the amount of software we have today, and I don't believe we are. I'm certainly not. I've always had a list of ideas that were never worth the effort, and now suddenly they are. And I've rarely seen a backlog at work that was actually empty.
If it becomes a hundred times easier to build software, I don't think we'll build the same amount and lean back. We'll just build a lot more.
DHH made a similar point in his Rails World 2026 keynote about the end of programming as we knew it. He says he hasn't written any code in about five months, and yet every dream, every annoyance and every "frivolous feature" is "suddenly within reach." Later in the talk he brings up the introduction of the ATM: people expected bank tellers to lose their jobs, but because running a branch got cheaper, banks opened more branches and hired more tellers.
I recognise the first part a lot. I'm less sure about the second. My own experience shows that there is a lot more work to do, but that doesn't mean there will be more paid jobs. Those are different things. Someone still has to want all that extra software badly enough to pay for it, and a lot of what I build now is stuff nobody would have paid me for before. What I do know is that the number of things that are worth building has exploded now that building them has become so much cheaper and faster.
Where my time goes now
I used to spend eight hours programming a single feature. Now I'm talking to different agents all day while they work on all kinds of projects in parallel. I come up with new ideas, improve existing products, point agents in the right direction, judge whether what they made is actually useful, and add my own taste to it.
The bottleneck is no longer how fast I can write code. It's how many good decisions I can make in a day and how much work I can keep track of. There's no natural limit to that, which is exactly why my days are so full.
A big part of it is the human side. An agent that doesn't get enough context can't magically figure out what a customer really needs. It's also not always good at judging what feels intuitive or logical to a person. So I spend a lot of my time talking to people, trying to understand what they mean, and turning that into something agents can work with. Technically correct software is easy now. Software that is actually valuable to someone still needs a person who understands that someone.
This won't last either
I don't think this is permanent. Agents are getting better at product decisions too, and at some point they may become better product owners than I am. Given how quickly the previous steps went, it would be naive to think the current division of work between me and my agents is final. I've been wrong about this kind of thing before: advice I gave on this blog in February was outdated a few months later.
I see two ways this can go. The first is that the same thing happens again: my role shifts once more, and we find something new to do on top of whatever the agents have taken over. That's what has happened every time so far. The second is that AI eventually really can do everything. Then we may end up in some kind of utopia where nobody has to work anymore, and whether there's still work for programmers isn't a very interesting question at that point.
I have no idea which one it's going to be. Like DHH said, no one knows anything about the future.
For now
I understand the fear that AI will make programmers redundant. I've had that feeling myself, more than once. But every time, there turned out to be a new layer of work on top. First I wrote code. Then I reviewed code. Then I guarded architecture and quality. Now I mostly work on product development, user experience and steering agents.
My work hasn't disappeared, it has changed. And because I can get so much more done now, I actually have more to do.
Maybe the biggest mistake in the discussion about AI and jobs is assuming that the amount of work that needs to be done stays the same. That's at least not what I'm seeing in my own work.