Think of two managers in the same team in Singapore. Both use the same AI assistant. Both send their director a recommendation on the same supplier contract. The two papers look almost the same. They have the same structure, the same figures and the same clear conclusion.
One manager read the AI's draft, then opened the contract. She checked two figures against the finance system. She called the account lead in Jakarta to ask about a late delivery. She changed one line. The other manager read the draft, liked it and sent it.
The director sees two good papers. The difference between the two managers is invisible. It will only show later, in the work, if the AI was wrong about something.
The answer is no longer the hard part
For most of working life, the hard part of a decision was finding an answer. You collected data, built a model and wrote a paper. Good people were the ones who could do that work well and quickly.
AI has changed this. An assistant can now summarise a report, draft a proposal and suggest a conclusion in seconds. Many people in a team can now get a similar draft quickly. When the drafts look alike, the draft tells you less about the person who sent it.
What still differs is what each person does next. Do they open the source? Do they ask a colleague with different facts? Do they push back on a claim that sounds too neat? Do they accept a sound answer quickly, without wasting a day? These habits now decide the quality of what a team does with AI.
The pair can lose what each side brings
On average, AI does help. A large review of 106 experiments found that people with AI beat people working alone. That is good news, and it is why organisations are right to roll AI out.
The same review found something else. The person and the AI together often did worse than the better of the two working alone. The gap was largest on decision tasks. In other words, the pair can lose some of what each side brings. The way the person works with the AI decides how much is lost.
This is the part of an AI rollout that few people are looking at yet. The tools are bought. People are clearly faster. The quality of what they decide with AI is much harder to see.
Habits, not tools
Each person forms quiet habits with their AI assistant. Some people question every answer. Some accept an answer because it sounded right. Some ignore the AI even when it is correct, because they do not trust it.
These habits matter more than they seem. When AI is part of every workflow, personal habits add up. They become how a department checks a forecast, approves a payment or answers a customer. They become how the organisation behaves.
The scale is already large. In a 2025 study, KPMG and the University of Melbourne surveyed more than 48,000 people in 47 countries. Of the employees who use AI at work, 66% said they had relied on AI output without evaluating its accuracy at least once. Of the same employees, 42% said they did so sometimes or more often. That does not mean the output was wrong.
What a leader can ask
A usage dashboard tells you how often people open the tool. A training record tells you who sat the course. Neither tells you what happened between the answer and the decision.
So the useful question for a leader is not "Are my people using AI?" It is closer to this: when my people decide with AI, what do they check, and what do they let through?
You can start without any new system. In your next review of an AI-assisted paper, ask three simple questions.
- What did the AI get wrong, or leave out?
- What did you check yourself, and how?
- What would have changed your mind?
If the answers are quick and specific, the person was in charge of the decision. If the answers are vague, the AI may have been in charge.
Sources
- Vaccaro, Almaatouq and Malone (2024). Systematic review and meta-analysis, Nature Human Behaviour. arXiv.
- KPMG and the University of Melbourne (2025). Trust, attitudes and use of AI: A global study 2025. Survey of 48,000+ people in 47 countries, November 2024 to January 2025. Press release. Country insights.
