AI is not creating a two-tier workforce. It is revealing one that already exists – and accelerating the distance between the tiers. The decisions being made in organisations right now will determine whether that divide becomes permanent. People leaders are uniquely positioned to influence those decisions. Most are not yet in the room where they are made.

“2026 technology on a 1996 business model. We’re layering AI onto archaic processes without redesigning the end-to-end user journey.” 

“99% of the workforce is in a GenAI frenzy focused only on language models. A small fraction understand deeper capability. That gap is where the value gets lost.” 

“It’s not about being behind the curve. Even the biggest players aren’t sure what’s next. The advantage isn’t speed – it’s being deliberate.” 

“AI is being dropped on top of the day job. We’re not stopping to ask how work could be fundamentally better.” 

“There is no curve to be behind. That framing itself is part of the problem.”

42% of employees expect their role to change significantly due to AI within the next year. Yet only 17% use AI frequently today, and 34% feel unprepared. Most telling: 42% say their employer expects them to learn AI on their own. Organisations with structured, workforce-wide AI upskilling are nearly twice as likely to report significant ROI. Yet only 35% have such a programme in place.

One participant introduced a framing that has stayed with us: the talent pipeline is no longer shaped like a pyramid. It is becoming a barrel. Decades of under-investment in mid-level development, combined with the automation of entry-level roles and the reluctance to hire juniors, is creating a workforce that is broad in the middle and hollow at both ends. There are no apprentices learning the craft. There are no senior leaders who came up through the new kind of work. There is just the middle, holding everything together.

A second observation was equally sharp: AI is being deployed as a conversation, not a change. Tools are handed to people on top of their existing workloads. Nobody is asking the more fundamental question – how could this work be redesigned? That distinction, between AI as addition and AI as transformation, is where most organisations are currently stuck.

The knowledge gap is not evenly distributed. It tracks directly with seniority, income, and access. UK-specific research from Cornerstone OnDemand (2,000 UK employees, October 2025)^[12] is stark: only 16% of employees earning £15,000 or less receive formal AI training often or always, compared to 47% of those earning over £55,000 – a near-threefold gap. 51% of UK employees have never or rarely received any AI training or tools from their employer at all. The people closest to the work being transformed are the least equipped to shape how that transformation unfolds.

AI rollouts in most organisations have followed a familiar pattern: pilot with senior knowledge workers, expand to corporate functions, defer frontline and operational roles until the technology matures. The result is a widening gap between those who are learning to work with AI and those who are simply working harder without it.

Global research from Boston Consulting Group (2024)^[10] shows that eighty percent of the global workforce are deskless workers. They receive less than 1% of corporate technology budgets. Only 21% of HR functions are closely involved in AI strategy decisions. Meanwhile, 80% of UK employees have used AI to support their work – but 81% do not always tell their manager or colleagues they are doing so (Cornerstone OnDemand, November 2025)^[12]. Only 33% say they are often or always encouraged to use AI at work. 51% have never or rarely received any training or official tools to support AI use. Shadow AI is filling the gap that formal strategy has left open.

Global data indicates that frontline workers with AI access report a 41% burnout rate, compared to 54% for those without. Organisations that digitally connect 75% or more of their frontline workforce see 20% or greater bottom-line growth. The business case is clear. The investment pattern is not.

The answer, across sectors, is consistent: women in administrative and clerical roles. Older workers in mid-career transitions. Neurodivergent employees and colleagues with learning disabilities who currently hold the repetitive, structured roles most exposed to automation – roles that provided precisely the kind of predictable, contained work that enabled them to contribute and progress. These are not edge cases – they are the majority of the workforce in many functions, and they are disproportionately absent from AI governance conversations.

One participant named this as the second order impact: not the immediate disruption of AI adoption, but what happens after the first wave. Roles disappear. People are not redeployed. The question of who fills the resulting vacuum – and who is left without a pathway – is not being asked at the same speed as the automation decisions are being made.

Under UK law, liability does not sit with the software provider. It sits with the employer that decides to deploy the system. Once AI processes candidate data, ranks applications, or influences rejection decisions, UK GDPR, the Data Protection Act 2018, and the Equality Act 2010 are engaged. AI hiring tools have been flagged for screening out experienced candidates via graduation dates, employment gaps, and CV language that correlates with age – a direct Equality Act risk. Ageism complaints during recruitment jumped 133% in Q1 2025.

AI is increasingly used in performance management, surfacing patterns in feedback, goal attainment, and productivity metrics. If trained on biased historical data, these tools risk reinforcing inequities in promotion, pay progression, and development access – at scale, and with algorithmic consistency that human decision-making never achieved. We acknowledge that audited, well-governed AI can deliver fairer hiring outcomes than unassisted human decision-making – which is precisely why governance is the issue, not the technology itself. Research from the Harvard Business Review (2025)^[16] found that when AI is adopted without that governance, it does not just apply a definition of merit – it reshapes what merit means, locking in one version that reflects the characteristics of those who were previously hired.

The ILO’s March 2026 report^[13] found that 29% of female-dominated occupations are exposed to generative AI, compared with 16% of male-dominated roles. Among workers facing the hardest transition – those with limited savings, fewer transferable skills, and less access to retraining – 86% are women.

72% of S&P 500 companies now flag AI as a material risk in their public disclosures – up from just 12% in 2023. (The Conference Board and ESGAUGE: AI Risk Disclosures in the S&P 500, October 2025.) ^[22] Yet even the most technology-adjacent C-suite roles lack recognised AI capability: only 44% of CIOs are rated as AI-savvy by their own CEOs – and CHROs are rated significantly lower still.(Gartner: CEO and Senior Business Executive Survey, conducted June-November 2024, published May 2025. )^[18] 88% of organisations now use AI in at least one business function – up from 78% a year earlier. Yet only 21% are closely involved in AI strategy decisions at any meaningful level. In 2026, DEI success will be judged by pay equity, fair access to development, and transparent promotion pathways – not statements or workshops. As AI becomes more embedded, DEI must act as a safeguard preventing technology from reinforcing old inequalities.

62% of UK employers predicting AI-driven workforce reductions say clerical, junior managerial, and administrative roles are most at risk. These are often the entry points to organisations – and the talent pipeline for future leadership. When those roles disappear without replacement pathways, the leadership bench shrinks with them.

What happens to human capability when we outsource thinking? Participants described watching junior colleagues reach for AI tools before attempting to solve problems themselves, and the erosion of institutional knowledge as experienced staff leave and their expertise is assumed to exist in the system.

One participant offered a longer-term frame: AI will eventually become latent infrastructure, like electricity or WiFi – present everywhere, invisible, assumed. The anxiety we are experiencing now is the anxiety of transition. But the design choices made during this transition are not temporary. They will set the defaults that persist once the anxiety fades. Another participant raised a productive counter-challenge: perhaps the right response to intellectual decay is to let it happen first – to allow the pendulum to swing towards full AI dependency, learn from the cost of that, and then design the rebalancing deliberately. That is a brave institutional choice. Most organisations will not make it consciously. The question is whether they make it by accident.

The pattern is already visible in professional services. Harvey – an AI tool now deployed by most major UK law firms – writes and reviews contracts with approximately 99% accuracy. One managing partner described the consequence as an inverted pyramid: the firm no longer needs the volume of junior lawyers it once did. The barrel shape is coming. The question is whether organisations are designing for it or simply allowing it.

Gartner predicts that by end of 2026, 50% of global organisations will require AI-free skills assessments due to critical thinking atrophy. 57% of workers believe skill erosion – not job loss – will be AI’s biggest workforce issue in 2026. Among major technology companies – predominantly US-based – entry-level hiring has dropped from 15% to under 7% of total hiring. In the UK specifically, tech graduate vacancies fell 46% in 2024, with a projected further 53% drop by 2026.

If organisations stop hiring juniors in 2025, by 2030 they will face a shortage of the mid-level talent those juniors would have become. The roles being automated are often the roles where future leaders learned their craft. The efficiency gained today may be borrowed from the capability needed tomorrow.

One of the most striking observations from the roundtable did not come in the formal discussion. It came in the closing takeaways, when a participant from a cross-sector leadership role said simply: the panicking feeling that we have to perform AI competence is probably going to bite us.

It landed in the room because everyone recognised it. The knowledge gap in Fault Line 1 describes people not knowing enough about AI. This is different. This is about the social pressure to appear as though you do – to avoid looking behind the curve, to signal strategic credibility to boards and peers, to not be the leader who asks what might seem like a naive question.

The consequence is a specific kind of governance failure. Organisations procure AI tools to demonstrate commitment rather than to solve defined problems. Leaders endorse AI strategies they do not understand. HR and EDI leads, already excluded from decisions, find themselves unable to challenge technical claims they cannot interrogate. And at the other end of the spectrum, some senior leaders – aware they lack the knowledge to engage credibly – simply deny the relevance of AI altogether, creating a different version of the same problem.

Research from Gartner (2025)^[3] found that even CIOs – the most technology-proximate C-suite role – are rated as AI-savvy by only 44% of CEOs. For CHROs the confidence gap is wider still. That gap runs in both directions. People leaders who lack technical vocabulary cannot challenge vendor claims, audit procurement decisions, or interrogate the outputs of tools they are responsible for governing. The result is a function that is simultaneously excluded from AI strategy and unequipped to demand inclusion.

81% of UK CEOs name AI investment as their top priority for 2026. Yet only 8% of leaders have deployed agentic AI to a point where it is delivering value. The gap between stated commitment and operational reality is precisely the environment in which credibility performance thrives – and in which poor governance decisions accumulate undetected.

The credibility gap has a specific workforce equity consequence. When AI adoption is driven by the need to be seen to act rather than the need to solve defined problems, the tools chosen are rarely the ones that would serve the broadest workforce. Frontline access, equitable reskilling, and demographic impact assessments are not the features that get prioritised in a procurement process shaped by the desire to perform AI leadership. They are exactly the features that get deprioritised.

There is also a longer-term talent consequence. When leaders cannot distinguish between what AI genuinely does well and what it is claimed to do, they cannot protect the human capabilities that AI cannot replicate. They cannot have the conversation about where critical thinking must be preserved, because they cannot credibly engage with where it is being eroded. The performance of competence crowds out the practice of it.

The clearest consensus from the roundtable was not about a specific risk or recommendation. It was about posture. The organisations making the most progress on AI adoption – and on doing it equitably – were not the fastest movers. They were the most deliberate ones. They were asking: what problem are we actually solving? Who does this affect and how do we know? What would we need to stop doing to make space for this? That posture – curious, structured, human-centred – is what distinguishes AI strategy from AI theatre. Most organisations are currently staging the latter while believing they are doing the former. The EQUITAS™ framework is a tool for building genuine strategy into governance – not just intentions, and not just optics.

The actions below are aimed at People leaders with some freedom of movement. We recognise that many organisations – particularly in the NHS, local government, and heavily regulated sectors – operate under significant structural constraints: nationally determined job roles, legacy technology systems, limited procurement autonomy, and funding environments that make even incremental change difficult. If that is your context, the most important action may not be from the list below. It may be building the evidence base and the language to make the case for change upward – to national bodies, regulators, or funding authorities. This paper is designed to support that case.

A specialist leadership development consultancy in its 20th year. Our Tri-Level™ methodology has delivered 35%+ promotion rates and 98% recommendation scores across the NHS, financial services, local government, energy, and the charity sector.

Inclusive Leadership  •  Entrepreneurial Leadership  •  AI-Ready Leadership

Jenny Garrett OBE holds an OBE, is a Freeman of the Worshipful Company of HR Professionals, and is a graduate of MIT’s AI Strategy and Leadership Programme. She co-hosts the AI for Equity podcast and co-authors AI for Equity: Creating a More Equitable Society for All (Emerald Publishing, October 2026).

enquiries@jennygarrett.global  |   jennygarrettglobal.com
AI for Equity podcast: available on all major platforms