By Darya Mishyna, Principal Business Psychologist at Matthew Syed Consulting
What do Pixar’s “Braintrust” and New Zealand’s All Blacks have in common? Despite their different fields, both drive success not by individual brilliance, but by a culture where people with diverse skills and perspectives contribute ideas and challenge each other constructively¹˒².
Today, companies like Amazon and Infosys are using the same principle to boost the success of their AI implementation efforts³˒⁴.
At the heart of this is psychological safety⁵˒⁶ – an environment where people feel empowered and responsible to contribute unique perspectives, collaborate with others, challenge the status quo, and learn from experience.
While academic research is still catching up with the frenetic pace of AI adoption, we believe the importance of fostering psychological safety is amplified, not diminished.
The data we have so far is striking. A research paper by MIT Technology Review found that leaders overwhelmingly view psychological safety as a driver of AI adoption, with 83% observing connections between psychological safety and tangible AI outcomes. And yet fewer than four in ten describe their organisation as having “very high” levels of psychological safety³.
Even more revealing, while 91% of employees say that it’s important to speak up, a sizable 61% admit they rarely do so in front of senior leaders⁸. Similarly, although experimentation-friendly companies have greater success with AI projects³, 41% of employees report that experimenting with new approaches is usually seen as a risk rather than opportunity in their organisation⁸.
In other words, the capability organisations most need is the one many currently lack.
So, what is really going on?
To answer that, we need to understand that AI challenges are not just technical problems. They are highly complex ones that involve trade-offs between accuracy and fairness, speed and risk, innovation and control. No single individual, no matter how talented, can see the full picture. Teams with diversity of thought – those that draw on differences in perspectives, insights, experiences and thinking styles – are able to explore a broader problem space. We call these teams “collectively intelligent”. However, collective intelligence on its own is not enough. It will remain latent unless psychological safety is also present.
From our research, we’ve identified four dynamics that emerge from the way teams handle diversity of thought and psychological safety. Only one of them creates the culture where employees can maximise the value from AI implementation.
1. Creative friction
This is the sweet spot, where diverse perspectives are brought together and actively challenged. The environment allows teams to test assumptions, discuss risks, mitigate bias, and improve outcomes rather than defaulting to consensus.
Example: A cross-functional AI team (data scientists, product managers, legal, HR, and operations) debate trade-offs between model accuracy, fairness, explainability, and business value, challenging each other to ensure the end solution is both effective and responsible.
2. Stagnation
Here, something subtle but dangerous occurs. The team appears aligned and decisions are made smoothly, but this might be a warning sign. Without diversity of thought or the safety to challenge decisions, teams fall into groupthink and conformity.
Example: An AI steering group with similar backgrounds pushes forward a model without fully questioning its limitations, biases or risks, resulting in poor adoption and limited business value.
3. Untapped potential
In this zone, diversity of thought exists but it’s muffled. The team may have diverse perspectives, but it’s not safe to speak up. People hold back dissenting views to avoid conflict or because they think it won’t matter.
Example: A cross-functional AI team includes diverse expertise and backgrounds, but junior or non-technical members hesitate to raise concerns, leading to overlooked risks and weaker outcomes.
4. Overconfidence trap
At first glance, this may look like a high-performing team. There is good energy and the team actively debates, but the perspectives are too similar. The group ends up reinforcing shared points of view rather than challenging them.
Example: An AI team made up primarily of data scientists and engineers rigorously debate model performance, but without input from business, legal, or user perspectives, they miss real-world risks and deploy a solution that struggles in practice.
It’s time to consider the human element of AI implementation
Here is the lesson from all of this: let’s not underestimate the human conditions required to make this incredible technology work. The challenge is often cultural rather than technical. The companies that unlock real value from AI will be those that create an environment where diverse thinking is expected, challenge is welcomed, and learning is continuous. And that is very human.
References
This article was first published in Learning Magazine Special Edition 2026. Click here to read the full edition.