The human layer: the hard part of working with AI

I have spent the past year rebuilding how I write software using AI tools. Learning to use the tools and adapting my workflow has been incredibly interesting and rewarding. Actually living and working with those tools has been emotionally draining.

Not enough people are talking about why the hardest part of adopting AI tools is in your own head.

I've spent the past year building a new AI-native software company. The technical side of that has been relentless. The tools and models change constantly. The methods that worked in March are outdated by September.

This is a familiar kind of hard - sure, it cycles much faster than 10 years ago, but engineers have always had to learn new tools. I know how to do that, and so do most of the people I've worked with. It takes time, but nobody is stuck.

What I didn't expect was how much the tools would change what a day's work feels like, and how mismatched my instincts were.

Feeling like a caveman

Every engineer carries a set of instincts built over years: how carefully to work, how long a task should take, when you have done enough for the day, how much of your week you can afford to spend reading and learning. Those instincts were trained over years of hard work in a world where typing and small administrative overhead was the bottleneck. Effort, hours and output felt linearly correlated, and we got used to that.

AI tools have broken that link: on some tasks I now produce five to ten times what I used to in an hour. On others, the ones that are limited by how fast I can think, very little has changed at all. Effort and output have come apart, and the old instincts no longer fit.

The problem is those instincts keep firing, and when an instinct fires in the wrong world it shows up as guilt, doubt, or a low-grade stress that follows you around all day.

I've lost count of how many developers, engineers and executives I've spoken to, who share this experience of being both the most productive they've ever been, but also the most ill at ease. Most notably, they often can't explain why.

That is the subject of this series, which I'm addressing across four posts. The first two are about instincts that are now wrong, the third is about the discipline that stops them firing, and the last is about what all of this means for the people you hire:

1. Brain fry. The instinct about how long a day is. Four hours with these tools empties the tank and produces what four days used to. Stepping away at lunchtime feels like skiving. It is the correct decision.
2. The learning ratio. The instinct about what to spend the week on. Learning used to be a luxury you squeezed in. Now the maths says it should be a third of your time, and the guilt about that is wrong.
3. The four modes of AI coding. The instinct about how carefully to work. There used to be one quality bar. Now there are four, and you have to pick before you start.
4. Delegate accountability, not tasks. The tools do the tasks. What a founder runs out of is accountability, so the people worth hiring are the ones who take ownership of an area. That favours self-aware, curious generalists over specialists in tasks that will not exist in two years.

Why it shows up in a small business first

I notice all of this because I mostly work alone. There are few colleagues to measure against, no team norms about when the day ends, no obvious measure of how much output is "enough".

But the same thing is happening inside larger organisations, more quietly. Engineers shipping prototype-grade code straight into products because nobody told them which bar applied. Teams learning less than they should because reading does not look like work. People burning out on days that would have been easy five years ago, and feeling lazy about it. The problems are the same. They are just harder to see when they are spread across hundreds of people.

The tools will keep getting better on their own, the limiting factor is how well the people can adapt.