Over the last stretch, I read seventeen of the major reports on human skills in the age of AI — McKinsey, BCG, Deloitte, the World Economic Forum, the OECD, MIT Sloan, Harvard Business Review, and others. Read together, they converge on a surprisingly consistent picture of what leaders and organisations should build.
The list is not wrong. But there is a layer beneath it that almost none of the reports names — and it is the layer that decides whether the skills on the list can be reached at all.
This piece, part of the ongoing series on breath and leadership, is about that missing layer: not which capacities matter in the AI era, but what determines whether we can actually access them.
The skills everyone now agrees on
Across the seventeen reports, the same human capacities keep surfacing as the ones that hold their value as AI absorbs more routine cognitive work:
Human skill | Why it holds value as AI advances |
|---|---|
AI literacy | Working fluently alongside the tools, not around them |
Critical judgment | Evaluating and correcting output, not just producing it |
Creative thinking | Generating what the models cannot |
Empathy and human connection | Reading and holding what happens between people |
Leadership and influence | Aligning and moving others through change |
Adaptability and continuous learning | Staying useful as the ground keeps shifting |
There is little serious disagreement here. This is the emerging consensus, and it is a reasonable one.
The gaps they identify in organisations
The reports are also candid that most organisations are not capturing the value they expected.
McKinsey finds that nearly all companies have invested in AI, yet only a minority report meaningful financial returns at the enterprise level. BCG makes it starker: a small share of organisations — around five percent — have captured substantial gains, and that same group shows roughly four times higher shareholder returns than everyone else.
The gap between investment and return is not mainly a technology gap. It is a human one. The reports increasingly point to the same conclusion: the organisations winning with AI are not better at the tools. They are better at the human side.
Two capacities the reports underweight
Reading closely, two human capacities show up around the edges of these reports but are rarely given the weight they deserve.
Self-awareness and metacognition. The ability to notice our own state and the quality of our own thinking in real time. As machines take over more of the output, knowing when our judgment is degraded — and being able to do something about it — becomes a frontline skill, not a soft one.
Ethical and moral discernment. As AI surfaces more complex value trade-offs and acts more autonomously on our behalf, discerning what should matter, when almost anything can be optimised, becomes foundational. The reports tend to file this under governance. It is closer to a core human capacity.
The deeper blind spot beneath the whole list
Here is the part almost none of the reports address.
Every skill on the consensus list — judgment, creativity, empathy, adaptability — is state-dependent. It is not a fixed possession we either have or lack. It is a capacity we have more or less access to depending on the condition of our nervous system in the moment.
The mechanism is well established. Executive functions — working memory, cognitive flexibility, inhibitory control — are supported by the prefrontal cortex. Under chronic or acute stress, working memory shrinks and cognitive flexibility narrows. Judgment goes with them: we cannot weigh nuance when the threat-response system is running the show.
The reports treat these skills as things to train and then possess. But a well-trained capacity a leader cannot reach under pressure is not much use in the moment that actually calls for it. Access is the variable underneath the list.
The stress epidemic AI is accelerating
This would matter at any time. It matters more now, because the baseline the workforce is starting from is already strained — and the way AI is being deployed is adding to the load rather than relieving it.
Start with the baseline:
Seventy-two percent of US workers report moderate-to-high stress — the highest reading in seven years.
Employee thriving has fallen sharply in a short span, to levels lower than during the pandemic.
The pipeline is not recovering: Gen Z is burning out sooner than any generation before, with peak burnout arriving strikingly early in their careers.
Onto that baseline, AI is layering a specific and measurable additional load. BCG's work on what they call "brain fry" identifies the mechanism directly: using AI in oversight mode — monitoring, correcting, and managing multiple tools at once — increases cognitive load and degrades the very skills the reports say will matter most.
The pattern in that research is worth stating plainly:
Frequent AI users report a meaningfully higher burnout rate than non-users.
A share of AI users already experience cognitive overload from AI oversight, higher still in some functions such as marketing.
Those workers make substantially more major errors and report markedly more decision fatigue.
Productivity peaks at a small number of simultaneous AI tools; beyond that, adding more AI makes performance worse.
The workforce was already stressed, and the way AI is being deployed is adding to the load rather than relieving it. The tools depend on human capacities that the same deployment is quietly eroding.
Where to start
If the missing variable is access — whether the skills on everyone's list are reachable under load — then the work is to build and protect that access. Four moves matter more than any single training programme.
Train for a higher baseline. Regulation is not only about recovering faster after stress. It is about raising the floor: a higher set-point that gives more resilience — absorbing challenge without losing access to our best thinking — and more range, the capacity to operate across a wider spectrum of demands while staying regulated. This is built over weeks and months through deliberate practice.
Integrate regulation into the workday. Short, structured breath-based practices at the start of meetings, before high-stakes decisions, and between periods of AI oversight. Kept brief, accessible, and grounded in physiology rather than wellness language. The goal is a performance infrastructure that keeps prefrontal function online.
Redesign the context so it supports regulation rather than eroding it. Meeting density, the ratio of oversight-mode AI work to recovery, the implicit signals a culture sends about availability and rest. These design decisions are as consequential as any training, because a well-regulated nervous system in a dysregulating environment still degrades.
Measure what the consultants missed. Most organisations track tool adoption and time saved. Neither captures whether the human capacities those tools depend on are actually accessible. Add cognitive-load assessment alongside productivity metrics, and track decision quality alongside throughput. What we cannot see eroding, we cannot protect.
The reports have the list right. What they leave out is that the small group of organisations pulling ahead are not better at technology. They are better at the human side — and the human side runs on a nervous system most organisations do not yet know they need to protect.
Sources & further reading
McKinsey & Company. The State of AI: Global Survey. https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai
Boston Consulting Group. Are You Generating Value from AI? The Widening Gap. https://www.bcg.com/publications/2025/are-you-generating-value-from-ai-the-widening-gap
Boston Consulting Group. When Using AI Leads to "Brain Fry." https://www.bcg.com/news/5march2026-when-using-ai-leads-brain-fry
World Economic Forum. The Future of Jobs Report 2025. https://www.weforum.org/publications/the-future-of-jobs-report-2025/
Mercer. Global Talent Trends. https://www.mercer.com/insights/people-strategy/future-of-work/global-talent-trends/
Aflac. Aflac WorkForces Report. https://www.aflac.com/business/resources/aflac-workforces-report/default.aspx
Arnsten, A. F. T. (2015). Stress weakens prefrontal networks: molecular insults to higher cognition. Nature Neuroscience. https://doi.org/10.1038/nn.4087
Balban, M. Y., et al. (2023). Brief structured respiration practices enhance mood and reduce physiological arousal. Cell Reports Medicine. https://doi.org/10.1016/j.xcrm.2022.100895
Frequently Asked Questions
What human skills do the AI-era reports agree on?
Across the major reports — McKinsey, BCG, the World Economic Forum, the OECD, MIT Sloan, Harvard Business Review, and others — the recurring list is AI literacy, critical judgment, creative thinking, empathy and human connection, leadership and influence, and adaptability. There is little serious disagreement about the list itself.
What do the reports get wrong or leave out?
They treat these skills as fixed possessions to train and then own. But every skill on the list is state-dependent — we have more or less access to it depending on the condition of our nervous system in the moment. Under stress, the prefrontal cortex is less available, working memory shrinks, and cognitive flexibility narrows, so judgment and creativity degrade regardless of how well they were trained. Access, not acquisition, is the missing variable.
Is AI making workplace stress worse?
The evidence points that way. The baseline is already strained — a large majority of US workers report moderate-to-high stress, the highest in years. BCG’s “brain fry” research shows that using AI in oversight mode, managing and correcting multiple tools, increases cognitive load and degrades the very skills the reports say matter most, with frequent AI users reporting higher burnout and more decision fatigue.
Why are only a small share of companies capturing AI value?
McKinsey finds nearly all companies have invested in AI but only a minority report meaningful returns, and BCG finds around five percent capturing substantial gains, with that group showing roughly four times higher shareholder returns. The differentiator is not technology; it is the human side — whether the capacities the tools depend on remain accessible.
Where should leaders and organisations start?
Four moves: train for a higher baseline through regular practice; integrate short, physiology-based regulation into the workday around meetings, decisions, and AI oversight; redesign the context — meeting density, recovery, cultural signals about rest — so it supports regulation rather than eroding it; and measure what most organisations miss, adding cognitive-load and decision-quality metrics alongside adoption and throughput.