There is a line you hear often. Tools cannot fix character. Give a brilliant tool to a lazy person, and you get a lazy person with a brilliant tool. What really happens is that the tool flatters the people already working hard. It explains a disappointing result without asking anyone in the room to change.
Look at what the research found. Brynjolfsson, Li, and Raymond tracked more than five thousand customer support agents. Productivity rose fifteen percent on average. The least experienced agents improved in terms of speed and quality. The most experienced agents barely moved, and their quality even slipped a little.
Noy and Zhang ran the same kind of test on professional writing. Time spent on the actual work dropped by forty percent, while quality rose by eighteen percent. The gap between strong and weak writers got smaller, not bigger.
Dell’Acqua and his colleagues studied seven hundred and fifty-eight consultants at BCG. The ones who performed below average improved by forty-three percent. The ones who performed above average improved by seventeen percent.
So, the tool does change people. It changes what someone can produce on an ordinary Tuesday, and it helps the struggling ones most. But look closely at what those studies measured. A set task. A few weeks. The goal already agreed upon by someone else.
None of the studies measured an organisation over three years, where nothing is agreed upon and the goal is the very thing people fight about. The gap between those two things is where the real question lives.
What AI Actually Does to an Organisation
AI gives you more of whatever you already reward. In other words, AI gives an organisation more of what it actually rewards, measured by who gets promoted, whose project survives the budget round, and which behaviour is never punished.
If people are rewarded for looking busy, you now get more convincing busyness. More documents. More decks. All of it faster and better-looking than before. If meetings reward the appearance of preparation, you get flawless preparation about nothing. If nobody has ever been thanked for saying a project should stop, AI will help write a beautiful case for keeping it alive. If your culture punishes people who raise problems early, AI will help them write a more careful, more polished version of the message that does not raise the problem.
Essentially, the tool never asks whether any of that was worth doing. It picks up the signal you are already sending and makes more of it.
Disengagement is not a personality trait. Neither is caution. Neither is waiting to be told what to think. These are things people learn. Someone designed the appraisal process that taught people initiative is risky. Someone approved the third organisational restructure that taught people their work does not survive. Someone watched a colleague get punished for honesty, said nothing, and the whole team learned the lesson in about four seconds. Your people did not decide to become careful. They were taught. If AI is producing more caution in your organisation, caution is what you have been paying for.
What to Do Before You Scale Anything
The question is not whether your people deserve the technology. The question is smaller and far more useful. What does this place actually reward? There are three ways to find out:
- Look at the last ten people promoted. Not the criteria on paper. The ten actual names. What did they have in common? Whatever it was, you are about to get a great deal more of it.
- Look at the last project that should have been stopped and was not. Who raised it? What happened to them afterwards? How long did it take?
- Look at what happens to the second idea in the room after a senior person has already given the first. Follow it. In most places, it quietly dies.
Then fix what you find before you scale the tool. Not after. Putting AI into a broken reward system does not buy you time to sort the system out later. It makes the problem bigger, faster, with a licence fee attached.
People like to say that when skill becomes cheap, character becomes the thing that sets you apart. Be careful with that one as well. Companies mostly win on money, distribution, data, and position, not on virtue. Plenty of decent organisations have lost to better-placed ones.
Producing work has become cheap. Doing the wrong work costs exactly what it always did.
When a strategy document took three weeks, the three weeks did something useful. Someone had to decide the document was worth that much time. But that filter has gone. A convincing, well-argued, nicely formatted case for almost anything now takes an afternoon.
So the scarce thing is not the ability to produce. It is the judgement to decide what deserves producing, and the standing to say so when the answer is unwelcome.
Judgement is built, not hired. It grows when people are handed real decisions, allowed to get some wrong, and not destroyed when they do. It grows when the person who stopped the project is treated better than the person who ran it for two more years.
Most organisations never build it. Building it means that senior people have to be told they are wrong and take it well. No training course delivers that.
The Question Worth Asking
Every organisation is about to find out what it has been rewarding, whether it wants to know or not. AI does not create the answer. It publishes it, loudly, and faster than the usual explanations can be written.
So here is the question to think about before the next rollout.
If everything this organisation currently rewards arrived twice as fast and looked twice as good, would we be better off?
Answer that honestly, and you will know what to fix first.
Sources
- Erik Brynjolfsson, Danielle Li and Lindsey Raymond, “Generative AI at Work,” Quarterly Journal of Economics 140, no. 2 (2025): 889–942.
- Shakked Noy and Whitney Zhang, “Experimental Evidence on the Productivity Effects of Generative Artificial Intelligence,” Science 381, no. 6654 (2023): 187–192.
- Fabrizio Dell’Acqua et al., “Navigating the Jagged Technological Frontier: Field Experimental Evidence of the Effects of Artificial Intelligence on Knowledge Worker Productivity and Quality,” Organization Science (2025); earlier as Harvard Business School Working Paper 24–013.
- Steven Kerr, “On the Folly of Rewarding A, While Hoping for B,” Academy of Management Journal 18, no. 4 (1975): 769–783.
Preparing your organisation for what comes next?
We help leaders build the adaptability, skills and ways of working that a fast-changing environment demands.
Explore Change Management & Adoption