Building the Artificial Intelligence Future on Cracking Ground
- Rachel Woodroof
- 6 days ago
- 6 min read
A note on AI adoption, psychological safety, and what we owe our people.
There is a particular kind of organizational irony that is quickly becomeing visible as we step back to see the whole shape of it.
We are rolling out Artificial Intelligence at scale — in teams, in workflows, in the daily texture of how people do their work. We are asking people to experiment, to learn, to fail forward, to raise their hand when something isn't working. We are, in other words, asking for precisely the conditions that require psychological safety to exist.
And the research is now confirming what many people in organizations have been quietly feeling: AI adoption is eroding the very safety it depends on.
The contradiction at the centre
Psychological safety — the sense that it is safe to speak up, to take risks, to make mistakes without punishment or humiliation — is well established as the foundation of high-performing teams. It is what allows people to bring problems to the surface before they compound, to question a decision without fearing a relationship rupture, to admit uncertainty without losing credibility.
And it also turns out to be an essential prerequisite for successful AI transformation.
Forrester Research has made this case explicitly in recent findings:
You cannot successfully integrate AI without first building the conditions in which people feel safe to fail, question, and flag concerns. The experimentation that AI adoption requires — the tolerance for getting it wrong, the willingness to flag an error, the capacity to say I don't know how this works yet — cannot happen in a culture of silence.
And yet we now learn, AI adoption is actively generating silence.
I don't always get this nerdy but here's the validated scoop: A peer-reviewed structural equation modelling study published earlier this year (2026) in Humanities and Social Sciences Communications (Nature portfolio) has now confirmed what practitioners suspected: AI adoption has a significant negative impact on psychological safety, which in turn increases employee depression. This is no longer inference. This is a confirmed causal pathway.
According to the ManpowerGroup 2026 Global Talent Barometer — drawing on nearly 14,000 workers across 19 countries — regular AI usage jumped 13% globally, while confidence in using that technology fell by 18%. Sixty-one percent of US workers are currently languishing (University of Illinois, February 2026). A Radical Candor survey published in May 2026 — drawing on 600 US workers across all levels — found that 61% of employees regularly observe colleagues going silent rather than share a dissenting view. That figure goes well beyond AI-heavy contexts. It describes the baseline of most teams.
Unfortunately, this isn't describing a niche problem in high tech industries- it's the operating environment most organizations are currently working inside.
What is actually happening
When AI makes an error — and it does, regularly — trust becomes ambiguous. People begin to doubt not only the tool, but their own judgment about when to trust the tool, and then — more quietly, more corrosively — their colleagues' judgment too.
This "trust ambiguity" compounds. It creates a workplace where nobody is quite sure what is reliable. And when reliability is uncertain, people protect themselves. They go quiet. They self-edit, and they stop naming what they notice.
Meanwhile, organizations are accelerating AI rollouts - because the competitive pressure is real. They are asking their people to adopt faster, experiment more, and tolerate more uncertainty — without first attending to the conditions that make any of that possible. (Not Good)
There is also a structural layer. The OECD 2025 Skills Outlook found that 30% of roles in advanced economies are currently "structurally mismatched" with digital and AI work realities. We call this role incoherence — not knowing what your job actually is anymore, or how to do it well — which happens to be a primary burnout driver. Psychological safety challenges are often downstream of structural design failures, not just cultural ones. When people do not know what is expected of them, safety erodes at the foundation.
artificial intelligence Intelligence
What the best organizations are getting right
It's not a uniformly bleak picture.
In the Deloitte 2026 Gen Z and Millennial Survey — drawing on more than 22,500 respondents across 44 countries — found that psychological safety has modestly improved at better-performing organisations. So, not everyone-declining; in fact some organisations are pulling ahead.
What distinguishes these organizations? The same peer-reviewed study cited above offers a finding worth sitting with: ethical leadership moderates the negative effect of AI adoption on psychological safety. Leaders who model transparency, who hold uncertainty honestly, who create protected space for experimentation and honest failure — these leaders partially buffer what AI adoption otherwise erodes. These are leaders with higher emotional intelligence skills.
Not a small finding! It means the solution is not primarily technical. It is human. It is relational. It lives in the quality of leadership and the daily practices of how people are heard, held, and trusted in teams.
What this asks of us
Organizations that will navigate AI adoption well are not necessarily the ones moving fastest. They are the ones that have attended carefully to the human ground beneath their transformation efforts.
That looks like several things in practice:
Role clarity before AI rollout. When people understand what they are responsible for and what success looks like, they have a stable enough foundation to absorb change. Psychological safety cannot be built on structural confusion - so thinking ahead about roles and adjustments that may be coming sets a crucial clarity map.
Leaders who model vulnerability around uncertainty. One of the hardest and most attractive leadership skills. The question is not whether leaders know how to use AI — many don't, yet, and that is fine. The question is whether they can say so openly, demonstrate learning in real time, and resist the pressure to perform confidence they do not have. That modelling changes what is safe to say in a team.
Protected experimentation space. Not as a platitude, but as a design question: what actually happens when someone uses AI badly, flags an error, or admits they don't understand something? If the answer is silence, or a quiet note in a performance review, the space is not safe. The response design has to be honest in regards to psychological safety.
Measurement that accounts for downstream change. Many organisations currently measure safety-building interventions by session satisfaction. In an environment of systemic trust collapse, this won't keep working. The question isn't whether people felt good in the room — it is whether the conditions in the room changed what was possible outside it. Following up on EQ development results and their effect on psychological safety is key.
Investment in the development of human potential. Last, but definitely not least: this is the long view, and it's urgent. AI will continue to evolve faster than most organisations can manage. What will not become automated, and will refuse to become obsolete is the human capacity for judgment, attunement, relational trust, and adaptive thinking. These are not soft skills. They are the skills that make every other skill workable.
The Radical Candor survey found that when workers were asked what skills matter most in an AI-driven world, the vast majority named human qualities — building clarity and alignment, coaching and developing people, genuinely caring about relationships. Technical know-how came in far behind. More than half of executives admitted they are underinvesting here.
In Radical Candor's survey they found that when workers were asked what skills matter most in an AI-driven world, the vast majority named human qualities — building clarity and alignment, coaching and developing people, genuinely caring about relationships. Technical know-how came in far behind. More than half of executives admitted they are underinvesting here. Seventy-six percent of workers have noticed. The ManpowerGroup data shows exactly where that gap lives: 56% of the global workforce received no recent training, and 57% had no access to mentorship. Organisations are accelerating AI adoption and at the same time withdrawing the developmental support that makes adoption sustainable. That isn't a technology problem. It's a human one.
A word about pacing
There is nothing inherently wrong with moving quickly. But speed without a foundation isn't efficiency — it is debt, accumulating interest in the form of burnout, disengagement, and the slow erosion of people's sense that their voice matters.
The organizations that will thrive in the next decade aren't the ones who adopted AI fastest. They're the ones who built human cultures strong enough to hold the weight of change — who invested in the relational infrastructure of their teams before the pressure demanded it.
That investment is available now. The conditions can be built. But they cannot be built in retrospect, after the damage is already done.
At Greenhouse Amsterdam, we work with leaders and teams on the interior conditions that make organizational change possible: emotional intelligence, attuned listening, and the practices of psychological safety. If your organisation is navigating AI adoption and wants to strengthen the human foundation beneath it, we would love to be in conversation.
We are genuinely curious about your thoughts on the topic of AI adoption and human development. What are you noticing in your own workplace? What is being asked of your people, and what is being made room for? The discourse matters and I hope to have some conversation in our comments section.
Wishing you peace and every good.
Rachel Woodroof is the founder of Greenhouse Amsterdam, where she works with organisations on emotional intelligence, attuned listening, and leadership culture.
Comments