Tedium has left the chat: AI, happiness and three ways to waste it
AI is the first technology in years that people say makes their work happier, and happier people measurably produce more. Most organisations are about to waste that in three ways: by counting it as headcount, by buying licences without teaching anyone, and by letting the thinking go with the tedium.
1. The work beneath us
There is a film I have watched more than once, Hector and the Search for Happiness, in which Simon Pegg plays a psychiatrist who travels the world trying to work out what makes people happy, writing each lesson in a notebook as he goes. I like it because it treats happiness as something worth defining carefully. Work has taught me to be suspicious of any word we use constantly and never define, and “productivity” is the worst offender I know.
A good part of happiness at work is the deletion of the work you feel is beneath you.
I don’t mean beneath your grade. In procurement I mean chasing approvals, re-keying supplier details and answering “where is my PO?” for the fourth time that week. Everywhere else it is the formatting, the copying of numbers from one system into another, and the meeting notes nobody will read: the hours you spend knowing that none of it is why anyone hired you.
One of my first jobs was in a post room, at a time when large companies still kept whole teams for the things an office needed doing. Those teams have gone and nobody mourns the queue, because the work moved to the person who needed it done, and the tools made it trivial. That is the pattern I have watched for thirty years: work migrates from a specialist pool to whoever has the need, and then the tool eats the boring part of it.
I went on to start my career as a software engineer, and developers have one habit that has never changed. When something annoys them, they build a tool to make it stop, which means they live with every new way of working a few years before the rest of us. So when I want to know what is about to happen to procurement, HR, finance or legal, I look at what has just happened to developers.
2. A solution looking for a problem
I remember when blockchain was going to change procurement, with every supplier on a ledger and every contract executing itself. In May 2019 Gartner predicted that 90% of blockchain-based supply chain initiatives would suffer “blockchain fatigue” by 2023, for want of strong use cases.1
I am an optimist about technology, and even I struggled to name one person whose working day blockchain had made happier. It was a solution looking for a problem, and the people selling it spent years running workshops to find one.
AI arrived the right way round, as a problem looking for a solution, and the problem was the boring part of almost everyone’s job.
Nobody needed a working group to discover that people dislike writing up meeting notes.
3. The developers went first
In 2022 GitHub surveyed more than 2,000 developers about its Copilot assistant, and Stanford published the results in its 2023 AI Index.2 Of those developers, 88% said they felt more productive and 96% said they got through repetitive tasks faster. Those are the numbers that made the headlines, and they are the less interesting half.
The other half is about how people felt.3
A senior software engineer in the study described what happens when the routine thinking goes to the machine: the thinking that is left is the fun part. In that engineer’s words, it “sets off a little spark that makes coding more fun and more efficient.”3 I recognise it from my own years writing code, and I would put it like this.
Less brainpower wasted on the mundane means more room for real thinking, and real thinking is the fun stuff.
None of those survey questions is really about code. Faster with repetitive tasks, less frustrated, more time for the work you enjoy: that describes clearing an inbox, building a deck or chasing a supplier as well as it describes writing software.
A year later Microsoft asked early users of its office assistant much the same questions about ordinary knowledge work, and got much the same answers. Once they had tried it, 77% did not want to give it up.4
The office gains were a little smaller than the developers’ and pointed in exactly the same direction. Independent researchers have since found the same thing outside software. In a study of 5,000 customer support agents, an AI assistant raised the number of issues resolved per hour by 14%, with the biggest gains going to the newest staff, and it improved both customer sentiment and employee retention.5
These figures deserve some care. GitHub and Microsoft were surveying their own users, and most of what they measured is what people said. The speed claims are contested too: in an early-2025 trial by the research group METR, experienced developers were slower with AI tools while believing they were faster, a gap I come back to in section 8.6
The happiness finding has survived all of this far better than the speed finding. When GitHub repeated the exercise with Accenture in 2024, 90% of developers said they felt more fulfilled in their jobs.7 Whatever else we argue about, people like their work more when the tedium has left the chat.
4. Then they refused to go back
The most telling evidence comes from a study that failed. METR, the group behind that early-2025 trial, tried to repeat it with late-2025 tools. It concluded that developers are probably faster with AI now, and that it could not measure by how much, for a reason that says more than the number would have.6
They were paid to work without AI. Many of them said no.
The reason was that developers no longer wanted to take part. METR reports a significant increase in developers declining to join the study because they did not wish to work without AI, even at $50 an hour for work on their own projects. It is careful to add that it had cut the rate from $150, which will have played a part.6
“My head’s going to explode if I try to do too much the old fashioned way because it’s like trying to get across the city walking when all of a sudden I was more used to taking an Uber.”
A survey tells you what people say, and this tells you what they do. Developers turned down paid work sooner than go back to doing it the old way, which is the strongest evidence in this article that the change is one people want. Nobody has ever turned down $50 an hour to hold on to their expenses system.
5. Happiness is a performance measure
In most business cases, “productivity” is a polite word for cost. It means the same output from fewer people, and it tells you nothing about whether anyone wanted to produce it. Procurement knows this definition better than any other function, having been measured on savings for thirty years. I think that is a thin and slightly sad definition, and the evidence says it also leaves money on the table.
The best study I know on this happens to come from a former employer of mine. Researchers from Oxford’s Saïd Business School, MIT and Erasmus University followed BT’s contact centre staff for six months, asking them each week how they felt by email, with five emoji buttons running from very sad to very happy. In the weeks when people were happy they were 13% more productive: they handled more calls per hour and converted more of them into sales, without working any more hours.8
The same study found that bad weather made the staff measurably less happy, which will surprise nobody who has worked in Britain.
Gallup has measured the same relationship at a much larger scale. Its analysis covers more than three million employees and compares the most engaged quarter of teams with the least engaged quarter inside the same companies.9
Then comes the uncomfortable number. Gallup’s latest global report puts employee engagement at 20% in 2025, the lowest since 2020, and estimates the cost of the other 80% at $10 trillion in lost productivity, about 9% of global GDP.10
Four in five people at work are not engaged, and we have just been handed a technology that most of its users say makes work more satisfying.
Put those findings side by side. Four in five people at work are not engaged, engagement is worth about a fifth on sales and profit, and we have just been handed a technology that large majorities of its users say makes their work more satisfying. That is the growth case for AI: it raises what I think of as the GDP of the company because people want to give more, and no restructuring is required to get it.
Most organisations already own the instrument that measures this, and they file it under HR. The annual engagement survey and GitHub’s Copilot survey ask very nearly the same questions.
6. The first waste: counting it as headcount
The quickest way to waste all of this is to put it into a business case as people.
The arithmetic is seductive. Thirty minutes a day saved by each of 160 people comes to 80 hours a day, which a spreadsheet will happily convert into ten full-time employees. Those ten people do not exist. They are a Frankenstein’s monster stitched together from 160 half-hours, and the test is simple: ask whose security pass you are going to collect.
Fragments of time are capacity, and capacity only becomes a saving if somebody deliberately does something else with it. When a business case counts fragments as headcount, two things follow. The saving never appears, so the CFO stops believing the next business case, and the people who were told a tool would make their work better hear instead that it was bought to make them fewer.
That second effect destroys the very thing the evidence says is valuable. The customer support study found that AI improved employee retention5, and nobody stays longer in a job because their time has been sold to finance in advance.
The better question for a business case is what the freed capacity is for. In procurement the answer is rarely mysterious: the supplier nobody has visited, the category strategy last refreshed three years ago, and the contract renewals that arrive as a surprise every single time.
7. The second waste: uneducated aspiration
There is a phrase I have grown very fond of: uneducated aspiration. It describes wanting the outcome without learning what the outcome takes, and it is the corporate equivalent of a gym membership bought in January.
BCG’s 2025 survey of more than 10,600 people in 11 countries and regions found that only a third of employees say they have been properly trained to use AI.11 More than three-quarters of leaders and managers use generative AI several times a week, while regular use among frontline employees has stalled at 51%. The people with the least tedium in their jobs are getting the most help removing it.
The same survey shows what closes the gap: leaders who visibly support the tools, and at least five hours of training with coaching in person.11 Only about a quarter of frontline employees say they get that support. Five hours is not much to ask, since most of us have sat through longer courses on how to lift a box.
This waste has procurement’s signature on it, quite literally. “Licences bought, nobody taught” is a purchasing failure, and procurement negotiates the deal. It can make training, coaching and adoption part of the contract and part of what the supplier gets paid for, which makes the second waste one that procurement can prevent for the whole business.
The studies in section 3 explain why this matters so much. In the customer support study, agents with two months’ experience and the assistant performed as well as agents with six months’ experience and no assistant.5 The largest gains sit with the newest and least confident people, and they are the ones a licence with no teaching leaves behind.
8. The third waste: atrophy of critical thinking
The third waste is the one that worries me, because it looks like success while it is happening.
Researchers at Microsoft and Carnegie Mellon surveyed 319 knowledge workers who use generative AI at least weekly, and collected 936 real examples of tasks they had used it for.12 The more confidence people had in the AI, the less critical thinking they applied to what it produced. The more confidence they had in their own skills, the more they checked, refined and challenged it. The authors warn that technologies used improperly “can and do result in the deterioration of cognitive faculties that ought to be preserved”. Citing earlier research on automation, they describe people deprived of routine practice being left “atrophied and unprepared” when the exceptions arise.12 That is what I mean by an atrophy of critical thinking.
Less critical thinking
People who trusted the tool more scrutinised its output less.
More critical thinking
People who trusted their own skills checked, refined and challenged it.
Developers have a protection here that the rest of us lack. Code either runs or it doesn’t, and a developer finds out within minutes whether the machine was right. A supplier email, a contract summary and a board paper have no unit tests. A meeting summary never throws an error, however wrong it is.
That is why METR’s early trial still matters, even with its speed result overtaken. The developers believed they were a fifth faster when they were a fifth slower, and METR’s follow-up repeats the warning that self-reported speed-ups “can be quite unreliable”6. If people who can test their output misjudge the tool by that much, the rest of us are flying on instruments we have never calibrated.
I am a believer by temperament, and what has kept me right more often than not is an engineer’s habit of asking what the numbers say before I repeat them. The assistant that takes the boring thinking will take the important thinking too, if we let it, and the minutes it saves are worth far less than the judgement.
AI makes a brilliant research assistant and a dangerously persuasive referee.
One organisation I have come across now treats a significant decision made without consulting AI as something close to professional malpractice. I like that more than I expected to, because it gets the relationship right. It makes AI another person in the room, and the point of a person in the room is that you can disagree with them.
9. What I would do about it
None of this needs a transformation programme. It needs four decisions, and procurement is well placed to make them, because writing down what good looks like and holding people to it is what the function does for a living.
Measure what the evidence measures
Put the developers’ questions into the engagement survey (more satisfying work, less frustration, less time on repetitive tasks) and report them next to quality, cycle time and retention, with hours saved as one line among several.
Give everyone the tools, and teach them
That means at least five hours each, coaching from someone who uses the tools well, and leaders who use them where their teams can see.
Decide what the freed time is for
Decide before it is freed, and write that into the business case in place of the phantom headcount.
Write the tests that knowledge work lacks
Agree which outputs a person checks against the source, which claims must cite their evidence, and which decisions need a second pair of eyes, and make challenging the machine part of the job description.
The last of those is the one I would start with, because it protects what the other three create.
Hector’s question is still the right one. If a whole layer of the job was work we never valued, it is worth asking what we thought the job was for underneath it. In procurement the answer has always been judgement: what is worth buying, what is worth the risk and who is worth trusting. People who spend their days on that are happier, and the evidence says they are also better at it.
The tedium was never the work. It was just the bit we hadn’t automated yet. 🙂
- Gartner: 90% of blockchain-based supply chain initiatives will suffer blockchain fatigue by 2023 (May 2019)
- Stanford AI Index Report 2023, chapter 4, figures 4.3.11 and 4.3.12, reporting GitHub’s 2022 developer survey and experiment
- GitHub: quantifying Copilot’s impact on developer productivity and happiness (2022)
- Microsoft Work Trend Index: what Copilot’s earliest users teach us (November 2023)
- Brynjolfsson, Li and Raymond: Generative AI at Work, and Stanford HAI’s summary
- METR: measuring the impact of early-2025 AI on experienced open-source developer productivity (July 2025), its February 2026 update, and The Register’s report
- GitHub and Accenture: Copilot’s impact in the enterprise (2024)
- University of Oxford: happy workers are 13% more productive (the BT contact centre study, 2019)
- Gallup Q12 meta-analysis, 11th edition
- Gallup State of the Global Workplace 2026
- BCG: AI at Work 2025
- Lee and others: The Impact of Generative AI on Critical Thinking (Microsoft Research and Carnegie Mellon, CHI 2025), reported by The Register and Campus Technology
© Adam Brown, 2026.

