For decades, researchers have been pretty consistent about what makes teams perform. J. Richard Hackman pointed to compelling direction, the right structure and a supportive environment; Amy Edmondson established the importance of psychological safety; Google spent a few years studying its own teams and reached a similar conclusion: how people work together matters enormously.
None of that is suddenly redundant because ChatGPT turned up. Something fundamental about the team itself has changed though: the capacity of its intelligence. Put more simply, the same five people can now bring considerably more brainpower to the problem.
The capacity of a team’s intelligence has changed
Historically, a team’s collective intelligence was largely bounded by the people within it, what they knew, what they’d experienced and how much information they could reasonably process. AI expands those boundaries. A team of five can now research, analyse, create and interrogate information at a scale that would previously have required considerably more people and time. Increasingly, they can also work alongside AI agents and draw on the accumulated intelligence of the company itself.
We’re already seeing what that looks like. AI-enabled engineers can move from problem to working code with considerably less of the production work that once sat between the two. That doesn’t make great engineers redundant; it shifts where their value sits. Architecture, judgement, product thinking and knowing whether what you’ve built is actually any good become more important as producing the code becomes easier.
The same is happening at leadership level. Executives can walk into a strategic discussion having already used AI to interrogate data, model scenarios and challenge an argument. Less time collectively discovering things everyone could have discovered beforehand, more time deciding what actually matters. Not a bad trade.
Early research points in the same direction. In a field experiment involving 776 professionals at Procter & Gamble, individuals working with AI performed at a level comparable to conventional two-person teams without AI on the product-development tasks studied.
The obvious interpretation is that one person can now do the work of two. I’d be careful with that. The more interesting implication is that the potential capacity of one person has changed. AI can give you 50 answers before breakfast; someone still has to work out whether any of them are good.
AI can give you 50 answers before breakfast; someone still has to work out whether any of them are good.
As the cost of answers falls, the value of the question rises
As the cost of producing answers falls, the value of asking the right question rises. Judgement, critical thinking, curiosity, emotional intelligence and strategic rigour aren’t shiny new AI-age skills. They’ve always separated exceptional people from everyone else. AI just makes the gap more expensive to ignore.
In other words, AI doesn’t make talent less important. It changes the return you can get from exceptional talent.
Build, buy, borrow and bot
It should change how we construct teams too. Historically, capability and headcount have been joined at the hip. Need more marketing capability, hire another marketer. Need more analysis, hire an analyst. A capability problem eventually became a recruitment problem.
It doesn’t have to anymore. A team might have an exceptional internal leader, bring in specialist expertise around a particular problem and use AI to absorb work that previously justified additional headcount. The team hasn’t necessarily become smaller; its boundaries have become less obvious.
Which is why we’ve started looking at teams through a slightly different lens at High Performance People: build, buy, borrow and bot. What capability should you develop, what talent is worth hiring, what expertise can you access when required, and what work really doesn’t need another salary attached to it? Start with the outcome, work backwards into the capabilities required, then decide where each should come from.
Otherwise, there’s a reasonable chance we’ll spend the next few years using AI to make yesterday’s organisations marginally faster.
Faster is not the same as better
Reports that took hours take minutes, research happens almost instantly and first drafts appear on demand. Useful, absolutely. But making individuals faster doesn’t automatically make the team better. A six-month field experiment involving just over 7,000 knowledge workers found that generative AI changed activities people could alter independently, but had little effect on coordination-heavy activities such as meetings.
There’s something wonderfully corporate about using artificial intelligence to save hours producing work and then spending those hours in exactly the same meetings. Which is why the bigger opportunity isn’t making the seven-step process faster. It’s asking why on earth it still has seven steps. Why does that meeting exist? Why does a decision travel through three layers? Why are highly capable people spending their time finding and summarising information machines can increasingly handle?
Today’s magic trick eventually becomes tomorrow’s Microsoft licence.
That doesn’t mean removing every human interaction. Some of the messiness of organisations is useful. It’s where trust develops, assumptions get challenged and ideas collide. The goal isn’t fewer humans; it’s fewer pointless interactions and more human attention where it creates disproportionate value.
What isn’t generic is the company using it
There’s one more part of this equation. AI itself won’t provide much of an advantage when everyone has access to increasingly similar AI. Today’s magic trick eventually becomes tomorrow’s Microsoft licence.
What isn’t generic is the company using it.
Every organisation has context no foundation model arrives with: its customers, data, intellectual property, decisions, commercial history and accumulated experience. Much of it is scattered across CRM records, strategy decks, customer conversations and the heads of a handful of people everybody knows to ask.
Connecting AI to that context means the next person tackling a problem doesn’t have to reconstruct what the organisation already knows. And when the questions asked, information retrieved, decisions made and outcomes become part of that shared intelligence, context starts to compound.
Instead of organisational knowledge disappearing into another PowerPoint deck nobody can find six months later, it becomes an intelligence layer that people and agents can contribute to and draw from.
So what should the team be made of?
None of this replaces what makes humans work well together. Trust, clarity, psychological safety, accountability and shared purpose still matter. Great people still matter enormously. AI hasn’t repealed human nature; it has changed the amount and type of intelligence we can put around those people.
For decades, the collective intelligence of a team was largely a function of who you put in it and how effectively they worked together. Increasingly, that team can combine permanent talent, specialist expertise, AI agents, automated workflows and the accumulated intelligence of the company itself.
Humans or AI is the wrong debate. The advantage is in how intelligently you combine them.
So one of the central questions in building a high-performing team is still who should be on it. But there’s now another one worth asking alongside it: what should the team be made of?
Humans or AI is the wrong debate. The advantage is in how intelligently you combine them.
Sources
- The Cybernetic Teammate: A Field Experiment on Generative AI Reshaping Teamwork and Expertise — Dell’Acqua, Ayoubi, Lifshitz-Assaf, Sadun, Mollick et al., NBER Working Paper 33641 (2025). https://www.nber.org/papers/w33641
- Shifting Work Patterns with Generative AI — Jaffe et al., Microsoft Research (2025). https://arxiv.org/abs/2504.11436