I have been building software for a long time now, and for most of that time the job looked pretty much the same. You sat down, you worked out what needed to be built, and then you typed. You typed for an hour or so, stopped to check what you had done, fixed the bits that were wrong, and then went back to typing.
When I look back at where the hours went, most of my productive time was spent with my fingers on the keyboard writing code. The reviewing happened in small doses along the way, a bit like tasting the sauce while you cook.
My day looks very different in 2026, and I think the change is bigger than most of us have stopped to notice.
The AI does the typing now
These days the AI does the typing for me. I write a prompt, which takes a few tens of seconds, the AI churns away for a little while, and then it hands me back a pile of work to look at. The churning part keeps getting shorter too as AI is getting faster, so the pile arrives much more quickly now than, say, in early 2026.
I used to spend most of my day typing with a little bit of reviewing, and now I spend most of my day reviewing with a little bit of typing. My job has become reading what the AI wrote, deciding whether it is right, and steering it towards the things that matter most in whatever we are building.
It sounds like a cushy arrangement, and in some ways it is, but it comes with a catch that I did not see coming.

Lots of little decisions, all day long
In the old world, an hour of typing was also an hour of rest from deciding. You had already made the decision about what to build, and the typing was just carrying it out. Then you would check your work, make a handful of calls about what to fix, and settle back in for another stretch at the keyboard.
Now the gap between decisions has collapsed from an hour to a minute or two. The AI comes back and wants to know whether this approach is fine, whether that naming is what I had in mind, and what should happen in the six edge cases it has just uncovered. AI in 2026 is very good at working in loops, so it can keep going on a task for a long time and come back with a lot of finished work, and every piece of it needs a yes, a no, or a “not quite, try it this way”.
Psychologists call this decision fatigue, which is the idea that the quality of your decisions drops off the more of them you make. I suspect plenty of programmers are feeling it without having a name for it.
You finish the day having typed almost nothing, and yet you are more worn out than you were when you wrote every line yourself. Nobody warned me that being the boss of a very fast junior developer would be this tiring.
So are we actually more productive?
The short answer is yes, because we are producing a lot more stuff than we used to. What surprised me is that all of that extra output has not turned into the huge jump in finished features you might expect.
You would expect that if the typing is nearly free, we would do the same work in a fraction of the time and knock off early. What I have found is that when something becomes cheap to produce, you end up producing more of it and the time you saved gets used up again.
So now that code is cheap, we produce more code, more tests, more documentation and more projects, and the day fills up just as it always did.
The documentation finally exists
Documentation is a good example of this. We have always known we should write it, and we have always had a very good excuse for not writing it, which is that the deadline was last week. So the feature shipped and the docs stayed on the list of things we would get to one day.
Now the documentation just turns up alongside the code, because asking for it costs me one extra sentence in a prompt. The coding part of the job is much quicker than it used to be, and I still have to review all of it, but what comes out the other end is a more complete piece of work than I would ever have delivered on my own.
So are we getting the jobs done faster? Yes, although the reviewing eats into the time the typing saves, and the jobs are being done properly now, which I count as a win.
The jobs I would never have started

The other side of producing more is the work that would simply never have happened. I recently did a full UI review of everything in HorseRecords, which is something I had wanted to do for about three years. I never got to it because I knew it would cost me days and days of programming time, and there was always something more urgent.
With AI, the whole review took half a day of going back and forth and making a lot of decisions, like what colour the buttons should be and which icons to use where. I can hardly claim I did that job faster, given that I was never going to do it at all.
Then there are the vibe-coded projects, like a little app I have that lets me take photos of my garden and turns them into a video diary of the plants as they grow. I would never have sat down and built that the old way, because no sensible person spends a week of evenings on a pumpkin documentary. Now it exists, and it makes me smile every time I use it. By the way, my pumpkins are going bunta this season!
What does that mean for a business?
For anyone running a business, this is worth thinking about. In total we are more productive, in the sense that far more is being produced, and I am releasing more features than I used to. My best guess is that it is somewhere between one and a half and two times as many, which is a long way short of the ten times you sometimes hear promised.
It does seem like a bit of a paradox, because I can create massive amounts of code very quickly and yet that does not translate into the same jump in functionality. A fair chunk of that code goes into handling edge cases, which AI is very good at spotting and which would have taken me ages to identify and code for on my own. The features themselves still need a human to decide what they are, review what was built and make all those little calls along the way, and that human still only has so many good decisions in them per day.
Each feature now arrives with documentation, with the edge cases thought through, and alongside a stack of improvements that would otherwise have sat on the wish list forever. The feature count has gone up a modest amount on the chart, and the product underneath it is in much better shape.
Engineering is about solving problems

I do not want to undersell the speed, though. A piece of coding that would have taken me a week now takes a day or two, and that still amazes me a little every time it happens. The reason it shows up as twice the features at best, and nowhere near five times, is that the days I save get spent on the documentation, the polish and the jobs that never used to make the list.
It took me a while to get comfortable with not writing the code myself, because for years the typing felt like the job. I have come around to the view that engineering is about solving problems, and how we get there matters a lot less than whether the problem gets solved. Nobody using HorseRecords cares whether I or a trained pony typed every line by hand/hoof or directed an AI to do it, as long as the thing works and their problem has gone away.
As for me, I am happier doing the job this way. I am more tired at the end of some days than I used to be, and I have had to learn to pace my decisions the way I once paced my typing, but I am building things I have wanted to build for years. If the price of that is a bit of decision fatigue, I will take the deal and go and have a look at how the garden is growing.

