The Capability Gap

Enterprise AI is entering a new phase. Last month we argued that the gap between AI ambition and AI execution is structural rather than technical: processes, data and infrastructure matter more than model selection. That is only half of the story. Once AI moves from pilot to production, a second constraint appears, and it is harder to solve. The binding constraint becomes the workforce.

The debate around whether AI replaces people is becoming less useful. The more relevant questions are who owns the quality of AI outputs, how existing teams adapt around AI, and what operating model can absorb the changes that follow.

Across the technology and telecommunications businesses we work with, we see a recurring pattern. Leadership teams initially treat AI adoption as a tooling decision. Six months later the discussion almost always shifts toward accountability, workflow redesign and skills.

Three themes consistently emerge:

• The capability gap inside organisations is now growing faster than the technology gap itself.
• The scarce resource in enterprise AI is not the AI engineer; it is the business expert who knows when AI is right and when it only sounds right.
• AI rarely creates restructuring decisions. More often, it accelerates decisions leadership teams were already facing.

The organisations that capture sustained value from AI will not necessarily be those that move fastest. They will be those that redesign their operating model around the people they already have, selectively hire the capabilities they cannot grow internally, and redefine what workplace literacy now means in 2026.

Context

The enterprise AI conversation has matured in stages. Two years ago, executives asked what AI could do. Twelve months ago, they asked whether it was real. Today, the question has shifted again: what does AI do to the people inside the business?

Last month’s edition argued that organisational readiness, process, data quality and infrastructure, is the constraint on AI moving from pilot to production. That argument has a quieter implication that most leadership teams have not yet engaged with: the same readiness must be built for people. The pattern is consistent across the technology and telecommunications businesses we work with. Capital is committed. Models are deployed. Productivity gains begin to materialise unevenly. And then the workforce conversation begins, usually without the operating-model decisions having been made in advance.

Five questions tend to surface at once. Can the existing workforce manage AI in production, or are new hires required? Is AI a threat to existing employees, or can they be redeployed? Where does AI pressure the workforce hardest? How is change management structured once AI is live? And what should the hiring process now look for? A sixth, less-asked but more important question runs through all five: who owns the quality of what AI produces once it is operating at scale?

The answers are increasingly visible in the market. They are not what most boards are preparing for.

Analysis

The workforce implications of enterprise AI are becoming visible through several recurring shifts. Together they reshape the operating model far more than the technology itself.

The capability gap, not just the adoption gap

The April edition identified the adoption gap, the distance between AI pilots and AI in production. That gap closes only when a deeper gap is also addressed: the capability gap inside the existing workforce. WEF and SHRM data converge on a striking figure. By 2030, 39% of core workplace skills will change, and skill demand is shifting 66% faster in AI-exposed roles than in those least exposed. In organisations that have already deployed AI, HR leaders report that 57% of the impact shows up as upskilling and reskilling, 39% as shifts in responsibilities, 24% as new roles, and only 7% as direct displacement.

The implication is clear. Enterprise AI is primarily a retraining problem, not a recruiting problem. Leadership teams that treat it as a recruiting problem end up bidding for scarce specialist talent in an overheated market, while the structural opportunity, building broad AI fluency across the workforce they already employ, goes unaddressed.

Upskill the many, hire the few

Most organisations instinctively respond to AI by hiring specialists. The larger opportunity is often the opposite.

Every organisation needs broad AI fluency across the workforce: people who can use AI tools critically, judge outputs and integrate them into daily work. This is taught, not hired.

Specialists still matter, but as a relatively thin layer around the organisation, data engineering, MLOps, model governance, AI risk and compliance, and the platform architecture that lets the rest of the business build safely. Targeted hiring and external partners pay off here. Across the workforce, what is needed is fluency, not expertise.

Most enterprises do not need to become AI companies. They need to become AI-ready organisations.

The business expert in the loop: who owns quality when AI is the author

Most organisations still have not answered a simple question: who owns quality once AI becomes the author?

The scarce resource is increasingly not the model or the AI engineer. It is the person inside the business who knows what a correct answer looks like. AI does not produce wrong answers; it produces plausible ones. Only people with deep institutional knowledge can reliably tell the difference.

Different organisations are assigning different titles to this role, analytics translator, AI product manager, business-AI translator, AI systems auditor, but the capability is consistent: deep domain expertise combined with enough AI fluency to frame problems, validate outputs and define what good looks like.

As AI adoption scales, organisations increasingly apply three levels of oversight:

• Human-in-the-loop for high-risk work
• Human-on-the-loop for monitored autonomy
• AI-supported review calibrated by humans for scale

In each case, technology scales judgement; it does not replace it.

AI does not produce wrong answers. It produces plausible ones. Only the people who understand the business can tell the difference, and they are the scarcest resource in the AI economy.

Redeployment is the real test of leadership

AI is increasingly appearing in restructuring discussions, particularly in customer support and routine knowledge work. Yet organisations reporting meaningful AI productivity gains have generally reinvested those gains rather than reduced workforce size.

AI rarely removes roles directly. It accelerates strategic decisions leadership teams were already facing.

The more important leadership question is therefore not headcount reduction but redeployment capacity. Roles often shrink faster than organisations can retrain people into new ones. Companies investing early in internal mobility, retraining pathways and cross-functional AI teams capture more value than those relying primarily on severance.

Where the workforce pressure lands hardest

AI pressure rarely arrives evenly across an organisation. The first impact usually appears in information-heavy and repetitive work: customer support, administrative processes, content production and routine knowledge tasks. Global call-centre labour savings alone are forecast at around $80 billion by 2026.

Operational-technology-heavy industries, telecom networks, manufacturing, utilities, logistics, face a slower but potentially larger shift, because AI redefines what the field and operations workforce actually does, not only how many of them there are.

For technology and telecommunications businesses, we see the largest exposure in contact centres and operational support functions. These areas are already being redesigned. The challenge is increasingly not external hiring but moving existing employees into new roles before capability gaps emerge.

Change management is the deployment

Technology implementation starts with tools. Human adoption starts with purpose.

Employees need to understand three things before training creates impact: why work is changing, what changes in their role, and who remains accountable.

Successful organisations typically build around four elements: a central AI function that sets standards, small multidisciplinary AI teams that deliver outcomes, clear executive accountability, and continuous change management rather than a one-off programme.

Two failure patterns repeatedly appear: too many pilots and too many conversations. One creates fragmentation. The other creates fatigue. Treating change management as a side activity almost always slows AI adoption.

Hiring for an AI-native workforce

The expectations baseline has quietly moved. AI literacy tops LinkedIn’s Skills on the Rise 2026 list, and healthcare, financial services, law firms and government contractors all now screen for AI proficiency in routine interviews. The old standard, comfortable with Word, Excel and PowerPoint, has been extended to comfortable applying AI copilots on top of them, and able to judge the outputs.

What employers actually screen for is not tool-naming. It is applied judgement. Can the candidate describe a real workflow where AI saved time, what they verified, and where it failed? The most sophisticated hirers now interview for the discipline to override AI as much as the fluency to use it. The job description, not just the interview, needs to catch up.

The enterprises that will capture sustained value from AI are not those that replace their people fastest. They are the ones that redeploy them most thoughtfully, and redesign their hiring bar so that the next generation walks in already AI-native.

Capvion perspective

In our experience advising companies, investors, and operators across technology and telecommunications markets, three principles distinguish the organisations that capture sustained value from enterprise AI from those that stall. None of them involve hiring more AI experts.

Make the business expert the star of the AI operating model

The most overlooked role in enterprise AI is not the data scientist. It is the domain expert who knows the company well enough to validate the AI’s outputs. McKinsey calls it the analytics translator; we call it the business expert in the loop. Build the operating model around them. Tier review intensity by risk: human-in-the-loop sign-off for high-stakes outputs, human-on-the-loop sampling for back-office autonomy, AI-on-AI evaluation calibrated against rubrics the domain expert wrote. The point is leverage. The expert’s job is to define what good looks like, not to check every output. Identify these people, name the role, protect the time, and put them where the accountability sits.

Build an internal mobility engine, not a severance programme

AI rarely fires people. It accelerates restructuring decisions leaders were already making. The honest boardroom conversation is about redeployment capacity, not absolute headcount. The organisations that win invest in internal mobility, skills marketplaces, structured retraining pathways, cross-functional AI pods that pull people from shrinking roles into growing ones. Severance is the lazy lever. Mobility is the strategic one. Get the data on internal skills before you need it, not after the cost programme is announced.

Rewrite the hiring bar before you need to hire

The 2026 baseline for office literacy is no longer Microsoft 365 fluency. It is AI judgement layered on top of it. Update the job description, then update the interview, then update onboarding and performance reviews. Screen for the discipline to override AI, not only the enthusiasm to use it. And remember that the hiring bar is a downstream consequence of the operating model. If the operating model treats AI as a productivity tool, hire for fluency. If it treats AI as an autonomous worker that needs supervision, hire for judgement. Most organisations are doing the second while interviewing for the first.

Conclusion

Last month we argued that the AI adoption gap is primarily an organisational readiness problem rather than a technology problem. The same logic now extends to people.

The capability gap inside enterprises is growing faster than the technology gap itself, and many leadership teams have not yet redesigned their operating model around that reality.

The path forward is straightforward, even if the execution is not. Start with the people already inside the organisation. Identify the domain experts who carry institutional knowledge and make them central to AI quality. Build broad AI fluency across the workforce. Hire selectively for capabilities that cannot be developed internally. Redesign the hiring bar around judgement rather than tool familiarity.

For decades, productivity meant asking people to execute processes faster. Increasingly it will mean asking people to supervise, challenge and improve machine-generated work.

The technology is real. The use cases are proven. Competitive pressure will continue to increase. The capability to absorb it is now the scarce resource, and that capability is built by people.