The AI Skills Gap: How Organizations and Leaders Are Closing It
Culture Shift: Closing the AI skills gap requires addressing cultural issues rather than just providing training.
Specific Mandates: Effective AI mandates need operational clarity, detailing workflows and accountability for successful adoption.
Tailored Learning: Role-specific training focused on immediate applications enhances AI adoption and behavioral change in teams.
Enablement Focus: AI should be viewed as an enablement function that improves efficiency rather than a replacement for existing roles.
Outcome Measurement: Assessing AI adoption must prioritize the quality of work produced instead of just usage metrics.
Across industries, organizations are grappling with the same uncomfortable reality: they have invested in AI tools, issued directives to use them, and watched AI adoption stall anyway.
The problem is not the technology. It is the gap between access and actual capability — and the growing recognition that closing it requires more than a mandate.
Unlock for Free
Create a free account to finish this piece and join a community of forward-thinking leaders unlocking tools, playbooks, and insights for thriving in the age of AI.
Have an account? Log In
Leaders on the front lines of this work are rethinking how skills get built, how adoption gets measured, and what it means to actually embed AI into the way teams work. To understand how organizations are tackling this in practice, I spoke with leaders across operations and project management about their approaches to closing this unprecedented gap in workforce readiness.
The Real Problem Is Culture, Not Just Training
The instinct to treat the AI skills gap as a knowledge problem is understandable, but practitioners say it misses the root cause. Moe Rosenfeld, CIO at eCopier Solutions, puts it directly: "Adoption is basically a culture problem dressed up as a training problem."
What looks like a lack of skills is often a lack of psychological safety, unclear expectations, or a leadership environment that sends contradictory signals. Rosenfeld continues: "The companies struggling with this are usually mandating AI from the top while making people nervous about using it. You can't do both." When adoption becomes a performance metric before it becomes a supported practice, people learn to look busy rather than learn to work differently.
Graham Mann, Founder, SEOTakeoff, reinforces this: "The AI knowledge gap gets worse when organizations treat it as tool training. The useful training is workflow training." Teaching someone how to use a platform is not the same as teaching them where it fits into how they do their job.
The AI knowledge gap gets worse when organizations treat it as tool training. The useful training is workflow training.
Why Vague AI Mandates Fail — and What Operational Ones Look Like
Even organizations with a genuine commitment to AI adoption often undermine themselves with directives too broad to execute. Mann warns against exactly this: "I would avoid blanket AI mandates like 'Everyone must use AI.' They create performative usage." In fact, Meta put a face to performative AI usage when it went viral recently for its leaked internal "token leaderboard."
When the standard is vague, people meet the letter of it without changing anything meaningful. Ilya Margolin, Strategic AI and Data Workflow Consultant, frames what a real mandate requires: "AI mandates need operational language. 'Use AI' is too vague. A serious mandate names the workflows, the allowed tools, the restricted data, the review standard, and the accountable owner." That kind of specificity turns a directive into something people can actually act on.
At the other end of the spectrum, Alexander Debelov, Founder and CEO, Go X, describes an organization that has stripped the mandate down to its clearest possible form: "the mandate is simple: if a task can be done by an agent, it should be." Where some organizations are still debating whether to adopt AI, others have made the default assumption that it will be used and are building from there.
AI mandates need operational language. 'Use AI' is too vague.
The danger in the middle ground — where there is encouragement without structure — is what Margolin flags as the quiet risk of unmanaged experimentation: "Hidden individual experimentation creates inconsistent quality and data exposure. Shared operating standards turn AI into a repeatable capability."
Role-Specific, Decision-Proximate Learning
One of the clearest patterns among leaders who have seen AI training succeed is that it is specific — specific to the role, specific to the decision, and specific to the workflow.
Stein Janssen, Chief Operating Officer, Poki, describes what this looks like in practice: "For example, a Project Manager's need-to-know is how to apply AI to scoping activities, handover processes, and issue tracking; while an Operations Lead's need-to-know is how to utilize AI for consistency throughout escalation process design."
Generic AI literacy programs may build awareness, but they rarely change behavior because they are not anchored to the problems people are actually trying to solve. Janssen is direct about why proximity matters: "This type of learning must be very close to where decisions are made because if it is too far removed from these decision-making processes, the learning will become theoretical and may result in no change in behavior."
The test of any training is whether it shows up in how someone works the following Monday — and that is far more likely when the training is built around the decisions they are already making.
How Organizations Are Structuring AI Training Programs
Organizations are taking different approaches to building AI capability at scale, and the most effective ones tend to layer foundational literacy with deeper, self-directed development.
Aniket Ghonge, Senior Supply Chain Manager at Amazon, describes the baseline: "At every level of the organization, they have some form of AI-based training. It's very basic training, such as explaining what the different types of AI are. This is available to everyone and is mandatory." That floor of shared understanding matters — it creates a common vocabulary and ensures that no one is navigating AI conversations from a complete standing start.
At every level of the organization, they have some form of AI-based training. It’s very basic training, such as explaining what the different types of AI are. This is available to everyone and is mandatory.
For those who want to go further, Amazon has built a path: "Amazon also started offering machine learning university certifications and courses for free to Amazonians," Ghonge notes. "So I'm actually enrolled in one of those courses, and it's a self-paced environment where you can actually learn about machine learning and different concepts."
At the team level, leaders are also building their own infrastructure. Jeff Chamberlain, Manager of Broadband Services and PMO at Frederick County Government, has taken a grassroots approach: "I've actually spun up a little AI task force on my team, a group of project managers globally who are really passionate about continuing to evolve our AI methods." These kinds of internal communities — people who care enough to push the work forward without being told to — tend to move faster and with more credibility than top-down programs alone.
I’ve actually spun up a little AI task force on my team, a group of project managers globally who are really passionate about continuing to evolve our AI methods.
AI as Enablement, Not Replacement — and What That Looks Like in Practice
One of the most consistent reframes coming from practitioners is a shift in how AI itself is categorized. Chamberlain articulates what his task force keeps returning to: "AI isn't a tool. It's not a method. It's an enablement function. At its core, it’s an enablement method to allow us to do things faster, better, and stronger. It is not meant to replace anything. It's meant to enable speed and quality. But that still requires people."
That framing matters because it changes what organizations are trying to build toward — not replacement, but amplification. Debelov offers a concrete example of what this looks like when the permission to use AI fully is actually granted. His organization deployed nine digital agents to make AI voice calls to more than 100 local businesses: "They [the agents] booked 15 partner meetings, and 15 of those became new partners on the ground. The total cost of the agent calls came to about $50. The last time we did the same outreach with humans, it cost us roughly $1,500 to book the same meeting volume."
The technology was not new. What was new was the decision to use it. Debelov's read on what is actually holding most organizations back is unambiguous: "The gap is not technical. It is permission."
Measuring Adoption the Right Way
If the goal is genuine capability rather than visible compliance, measurement has to reflect that. Margolin describes his own standard: "I measure AI adoption through the quality of work produced, not through the number of prompts or tool logins." Login counts and usage metrics are easy to game and tell you very little about whether AI is actually making work better.
I measure AI adoption through the quality of work produced, not through the number of prompts or tool logins.
Mann offers a similarly outcome-focused test: "Did the process get faster, clearer, or less error-prone without lowering quality? If not, the AI layer is probably just noise." The question is not whether AI was used. It is whether the work improved. Organizations that measure the former while claiming to care about the latter will keep producing the same result — surface-level adoption that never becomes a real capability.
Closing the Gap the Right Way
Closing the AI skills gap is not primarily a technical challenge or even a training challenge. It is a leadership challenge. The organizations making real progress are not the ones with the most sophisticated tools or the most aggressive mandates — they are the ones that have been honest about the culture conditions required for adoption to take hold, specific about what using AI actually means in each role, and disciplined about measuring outcomes rather than activity.
The gap is real, but so is the path forward. The decision to use AI has already been made. What remains is the harder work — building the culture, the specificity, and the measurement standards that turn that decision into something that actually sticks.Want more insights like these? Sign up for a free DPM account to hear from more experts like these.
