You can’t train your way out of the AI skills gap

Most enterprises I talk to say they have an AI skills gap. That sounds plausible right up until you look at what companies are doing. They are spending millions on copilots, launching AI academies, hiring chief AI officers and rolling out internal training at scale. Yet for all that activity, most organizations still do not move faster, decide better or operate in fundamentally new ways. That is the real tension at the center of enterprise AI right now: Companies think they have a skills problem, but what they really have is a work design problem. I have seen this pattern repeatedly. The organizations that get real value from AI are usually not the ones that train the fastest. They are the ones who redesign work sooner.

The AI skills gap is real, but it is not the whole story. In many enterprises, the bigger failure is that AI is being layered onto jobs, workflows and operating models built for a pre-AI world. People are learning new tools, then being sent back into the same meetings, approvals, handoffs and reporting structures that made work slow in the first place. Training may improve local productivity. It does not automatically redesign how the business runs.

That is why so many AI programs feel busy without becoming transformative. The organization can point to courses completed, licenses deployed and pilots launched, but the underlying system still behaves the same way. Decision latency stays high. Bottlenecks remain intact. Managers absorb more complexity, not less. Employees become faster in small ways while the enterprise remains slow in all the ways that matter.

This gap is showing up clearly in the research. Deloitte’s 2026 State of AI in the Enterprise report says insufficient worker skills are the biggest barrier to integrating AI into existing workflows. Yet the most common organizational response is education and reskilling, not role or workflow redesign. In fact, Deloitte explicitly notes that companies are much more focused on AI fluency than on re-architecting how work is done. The same tension appears in Wharton’s 2025 AI Adoption Report: Executive sponsorship is rising; chief AI officer roles are now present in 6 out of 10 enterprises and capability building is still falling short of ambition. The signal is hard to miss. Enterprises know AI matters. Many are investing. But they are still treating adoption as a learning problem when it is really an operating model problem.

Training creates users. Redesign creates advantage

Training matters. Every enterprise needs a baseline level of AI fluency. People need to understand where AI is strong, where it is weak and how to use it responsibly. They need to know how to challenge outputs, apply judgment and separate acceleration from automation. None of that is optional anymore.

But training alone does not create an enterprise advantage. At best, it creates pockets of local efficiency.

An individual contributor may draft faster. A manager may synthesize information faster. An analyst may produce a first pass in less time. Those gains are real, but they do not automatically translate into better operating performance. In many organizations, the efficiency never reaches the P&L. It gets trapped inside legacy workflows, approval layers, meeting culture and fragmented decision rights.

That is the real issue. AI may already be improving work at the individual level, while the enterprise itself remains structurally unchanged. The Writer enterprise AI adoption survey found that executives see AI super-users as at least five times more productive than their peers, yet only 29% of organizations report significant ROI from generative AI. The contrast is telling. The constraint is no longer whether employees can use AI. The constraint is whether the organization is designed to convert those gains into faster decisions, shorter cycle times, higher throughput and better business outcomes.

This is where many AI initiatives quietly stall. Leaders can point to adoption metrics, training completion rates and growing license counts. Employees can honestly say they are using the tools. But the business still does not feel materially more responsive. Revenue does not move faster. Product cycles do not compress enough. Decision latency remains high. Management complexity increases instead of falling.

That is why I believe the wrong question is, “How do we train our people on AI?”

The better question is, “Which work should humans continue to own, which work should AI accelerate and which workflows should be redesigned entirely now that AI exists?”

That is the shift that matters. It moves the conversation from individual capability to institutional performance. It moves AI from a training initiative to an operating model decision.

And that is where CIOs must lead. The organizations creating advantages with AI are not simply teaching employees new tools. They are redesigning roles, workflows and management systems so that individual productivity gains become enterprise-level outcomes. Companies do not fall behind because AI arrived. They fall behind because they kept the same work design after it did.

The real AI shift is separating judgment from execution

The most important AI transformations I have seen do not start with tools. They start with a harder leadership discipline: Separating judgment work from execution work.

Once AI can reliably handle portions of execution, the role itself must be reconsidered. Not eliminated. Reconsidered. The question is no longer just how to make people faster inside the job as it exists. The question is whether the job was designed correctly in the first place.

That is where the real work begins. Leaders must deconstruct work below the level of titles and org charts. This is a much harder challenge than deploying a copilot. It forces decisions about spans of control, management layers, performance expectations and career paths. It changes what excellence looks like across the enterprise. And it changes what companies should reward.

If AI takes on more drafting, synthesis, retrieval and coordination, then the value of human work moves up the stack. The ability to frame the problem, define quality and make accountable decisions becomes more valuable than manually producing every intermediate step.

This is also why so many employees feel uneasy even when leaders talk about AI in optimistic terms. They are being told to use new tools, but not what the organization will still need uniquely from them. They are hearing about productivity, but not about role evolution. Training without redesign does not feel like empowerment. It feels like a shot across the bow of their career.

That is why the most important workforce conversation in AI is not about tool usage. It is about role clarity. People need to understand where human judgment still creates value, where AI should accelerate execution and how their path to relevance and mastery changes as a result.

That is not an HR side discussion. It is one of the central leadership tasks of the AI era.

CIOs must lead the redesign, not just the rollout

This is where CIOs have a larger role than many companies still recognize.

AI adoption is often framed as a cross-functional initiative, and of course it is. But when AI moves from experimentation into execution, the CIO is often the only executive with a clear view of the full system: Workflows, dependencies, security, data architecture, control points, operational friction and how work moves across the enterprise. That perspective matters because the next phase of AI is not about individual productivity. It is about institutional redesign.

That means CIOs cannot limit their role to rollout, enablement and tool selection. They must help redesign how the business operates.

In my view, that starts with three questions:

1. Where is AI simply making existing work cheaper or faster, and where could it allow the business to operate differently? Those are not the same thing.
2. Which roles need to be rebuilt? Not renamed. Rebuilt. If analysts spend less time gathering information, what should the organization expect more of in return? If leadership cannot answer that, it is not redesigning work. It is just hoping individual productivity turns into enterprise value on its own.
3. What new management disciplines does AI require? As AI becomes part of execution, leaders need clearer standards for validation, accountability and quality control. AI can compress execution, but it can also multiply errors at scale. That raises the premium on operating discipline, not lowers it.

This is why I think the skills-gap narrative can be misleading. It invites leaders to believe the problem is mostly educational, as if enough courses, certifications and training hours will somehow carry the organization into the future. They will not. They are necessary, but they are nowhere near sufficient.

The companies that pull ahead will treat AI as a redesign moment. They will rethink work at the level of tasks, decisions, teams and operating models. They will create roles with more judgment and less administrative drag. They will redesign career paths, so people are not just trained on AI, but advanced through the responsible use of it. And they will measure success not only through adoption, but through decision velocity, throughput, exception rates and business outcomes.

Most of all, they will stop asking employees to bolt AI onto broken systems.

That is the real opportunity in front of CIOs. Not just to deploy the tools. Not just to sponsor training. It is to help redesign the enterprise around a new division of labor between humans and machines.

The AI skills gap is real. But education alone will not close it.

Only better work design will.

This article is published as part of the Foundry Expert Contributor Network.
Want to join?