Over the past year or so, I’ve found myself using AI tools more and more: ChatGPT, Claude, coding assistants, and other tools that help me turn rough ideas into something more usable.
I’ve had many days where AI has made me feel very productive.
Annoyingly productive.
The kind of productive where you go from a rough thought to a structured argument, then to a better version, then to a counterargument, then to a rewritten version, then to a list of things you’ve missed – all in the time it would previously have taken to stare at a blank page and make another coffee.
This is clearly useful.
But I’ve also noticed something else. It’s tiring.
Not physically tiring in the same way as manual labour – but it is cognitively tiring.
Used well, AI does not do the work for you.
It accelerates the first part of the process, giving you more time to focus on the higher-level work.
Keeping up with the machine
But it also speeds up the pace at which you have to think.
You are constantly judging the AI output:
Is that right? Is that useful? Is that too generic? Has it missed the point? Has it accepted a bad premise? Is this phrasing better? Is that actually true? Does this fit the client? Does this sound like me? Can I rely on this? What have I not asked?
If used well – you should also be judging what you are doing with the AI output:
Is this my thinking, or just generic filler? Am I using the tool to enhance clarity and aid communication, or am I letting it create word-salad? Is it suggesting alternatives I’d not considered – or have I outsourced my thinking to it?
The tool is fast.
The human still has to keep up. After all, the human needs to maintain ownership and accountability – if I put my name to it, I can’t hide behind “oh, the AI did that bit”.
The work has not disappeared
A lot of the discussion about AI and productivity seems to assume that if AI makes something quicker, the work has simply reduced. The assumption is often that AI is automatically an efficiency play.
Sometimes that is the case. If AI saves you from formatting a table, summarising a long document, or turning rough notes into a first draft, that can be a real saving.
The question is – what do you do with that saving? It’s tempting to just move on to the next thing. But sometimes the better use of the saved time is to improve the work: challenge the draft, sharpen the argument, check the assumptions, and avoid treating AI output as “good enough” simply because it arrived quickly.
When using AI with knowledge work, the work often moves:
- Less time writing the first version. More time deciding whether the first version is any good.
- Less time searching for a structure. More time deciding which structure actually fits.
- Less time producing options. More time choosing between them.
- Less time doing the slow mechanical part. More time doing the judgement-heavy part.
That can be a very good trade. In many cases it is exactly what we want.
But we should not pretend that judgement-heavy work is effortless.
Eight hours of this?
This is where I start to wonder about the standard working day.
The default assumption for full-time work is still broadly eight hours a day, five days a week. That feels normal because it is what generations of us have grown up with.
But this isn’t a law of nature. It’s a historical settlement.
The eight-hour day came from a particular set of industrial-era arguments about labour, rest, productivity, and fairness. Later, the five-day week became the norm. These patterns were designed around particular kinds of work, particular technologies, and particular assumptions about management.
The 40-hour week was a huge improvement on what came before. It reflected a different balance between labour, productivity, profitability, and leisure.
Those assumptions about the nature of work have been remarkably durable. But just because they are long-lived, doesn’t mean they are permanent.
AI may change the shape of some work quite significantly
I can be very productive with AI. But I cannot sustain that high-intensity, AI-assisted, judgement-heavy mode for eight hours straight.
Maybe some people can. Good luck to them. I can’t.
And I suspect I’m not alone.
Even before AI, few of us were 100% productive for 40 hours between 9 and 5, Monday to Friday. We had slow days. We had days when our brains weren’t firing on all cylinders, so we picked lower-friction tasks: reconciling expenses, tidying the inbox, sorting admin.
AI may remove the easy work, not just the unnecessary work. And the easy work often plays a useful role in the rhythm of a working day.
So what happens if AI automates some of the lower-friction work and leaves us with more of the judgement-heavy work — the very work we cannot sustain indefinitely?
We already challenged one assumption
Only a few years ago Covid challenged one old assumption very quickly: that office work had to happen in the office. Before 2020, plenty of organisations treated remote work as unusual, risky, or a special favour. You’d need to apply, you’d need support from your manager and sign-off from HR – with a list of stipulations and check-in and review points.
Then lockdown happened and suddenly everyone who could work from home did so.
Hybrid work has real challenges. Communication can fragment. Junior staff can miss out on learning by osmosis. Some things are better in person.
Managing people working different hours from you is harder than managing people sitting nearby at the same time. It takes different skills and more proactive management. Things that used to happen as a by-product of everyone being in the same building now need to be thought about more deliberately.
The new pattern is not perfect. But the old certainly was broken.
Since Covid it’s become much harder to claim that office work always had to happen in the office – even as some organisations try to reassert older patterns through return-to-office mandates.
AI may do something similar to another assumption: that hours worked are a useful proxy for productivity.
Hours are easy to count
One reason hours remain so important is that they are easy to see.
Someone is at their desk. Someone is online. Someone is in the office. Someone is in meetings. Someone is visibly busy.
That is reassuring.
It is also a weak proxy for whether useful work is happening.
This is part of what sits underneath some return-to-office debates. There are good arguments for spending time together in person. Culture, mentoring, trust, informal problem-solving — these things are real.
But there is also a less impressive argument hiding underneath some of it:
It is easier to feel that people are working when you can see them.
That is not the same as knowing whether good work is being done.
The harder question is:
If someone produces the right work, to the right standard, at the right time, in a sustainable way, how much should the number of hours matter?
That sounds simple. It isn’t.
Measuring output is hard. Lots of important work does not fit neatly into a dashboard. Bad targets can be just as damaging as bad time-tracking. If you measure the wrong thing, people will quite reasonably optimise for the wrong thing.
Goodhart’s law applies here: “When a measure becomes a target, it ceases to be a good measure”.
So I am not arguing for a simplistic “just measure outcomes” answer. That’s an easy answer – but fiendishly difficult to implement.
How do you decide what the outcome should be? How do you deal with the challenges of different staff operating at different levels? If a senior member of staff can do something in a day that takes a junior two days – does that just mean senior staff get more time off – or more work?
But I do think AI makes the old “hours equals productivity” assumption look even weaker than it already did.
Not everyone gets this choice
There is an obvious caveat here – this does not apply equally to all work.
Many jobs require physical presence. Many are shift-based. Many involve direct care, service, logistics, performance, maintenance, or operational cover. You cannot simply tell everyone to work asynchronously, from wherever they like, whenever they feel most productive.
Even within office work, flexibility is not evenly distributed.
Senior people often get more freedom than junior people. People with scarce skills get more freedom than people with less bargaining power. Some people like remote work. Some hate it. Some people need flexibility because of caring responsibilities. Others need the separation of leaving work behind at the end of the day.
So this is not a “future of work” utopia.
I don’t think AI means we all work three days a week and spend the rest of our time reading novels in the garden. Nice though that would be.
The more realistic question is who benefits from the productivity gains.
- Does the worker get more breathing room?
- Does the organisation just expect more output?
- Does the customer get a better service?
- Does quality improve?
Or do we just cram more cognitive load into the same number of hours?
The reverse-centaur problem
Cory Doctorow has used the phrase “reverse centaur” to describe the situation where the human ends up serving the machine, rather than the machine serving the human.
That is a real risk.
The good version of AI at work is that it removes drudgery, reduces friction, and helps people focus on judgement, creativity, relationships, and decisions.
The bad version is that it speeds everything up, generates more material to review, creates more things to check, and leaves the human responsible for making the system usable — while still being expected to work at the old pace, for the old hours, with higher output.
That is not liberation. That is just a faster treadmill.
So where does this go?
I don’t know where this ends up – and I’m hugely suspicious of anyone who claims they do.
But I do think AI should make us more willing to question some inherited assumptions about work.
Not because everything old is wrong.
Not because technology magically solves the hard human problems.
And not because every job can be redesigned around flexibility.
But because tools shape work. And when the tools change significantly, the shape of the work may need to change too.
The question should not be:
“How much more can we produce with AI?”
It should be:
- “What pace of work is sustainable?”
- “What still needs human judgement?”
- “How do we know whether good work is happening?”
- “Who benefits from the saved time?”
And maybe most importantly:
“If the machine is faster, does the human have to be faster too?”
A note on this post
Yes, I used AI while writing this.
Not in a “write me a blog post about the future of work” way, then copy and paste the result while making a cup of tea.
That would be “prompt and paste” AI slop.
I used AI in the way I’ve described above: to test the argument, challenge the structure, spot repetition, suggest alternatives, and refine what I was trying to say.
AI was my thinking partner. It wasn’t my thinker.
The final judgement, edits, mistakes, and mildly overused dashes are mine.
Matt Kirby 
Leave a comment