What Is the AI Learning Problem?

AI investment is moving faster than employee behavior. According to a 2024 Gartner survey of more than 3,000 managers, only 8% of employees use AI in ways that meaningfully improve their work, even as Gartner research finds 95% of CIOs expect significant value from AI investments. That gap is why the AI learning problem matters now.

If you are trying to turn AI deployment into measurable performance, this article explains what the AI learning problem is, why it is different from a simple training gap, and how organizations can address it through employee learning and change management. The core issue is not that employees lack access. It is that enterprise AI adoption depends on new habits, better workflow design, and support inside the work itself.

That point came through clearly in WalkMe’s October 2026 webinar Ready or Not: What Enterprise AI Is Asking of Your People. As Hila Segal put it, many companies already “have the licenses and they rolled out the tools,” but the open question is why that has not “turned into real competence yet.” J.P. Gownder of Forrester answered even more directly: “It’s not even primarily a training problem, it’s a learning problem.”

What is the AI learning problem?

The AI learning problem is the gap between deploying AI tools and helping employees learn how to use them effectively in real work. In enterprise environments, that gap appears when people have licenses, logins, and executive encouragement, yet still do not use AI consistently, confidently, or productively.

That matters because learning is not the same as training. In the webinar, Gownder said many organizations still rely on “one and done training,” where employees get “an hour or two of training” and are then expected to know how to use AI. His conclusion was blunt: “Nothing could be further from the truth.” Employees need to “learn new behaviors in context during workflows,” not just sit through a session and move on.

The distinction is practical. Training transfers information. Learning builds judgment, habit formation, and repeatable use in live tasks. Gownder’s Workforce AIQ framework reflects that broader challenge by measuring readiness across understanding, hard skills, soft skills, and ethics. He noted that only 22% of employees fall into the high AIQ segment, while 34% remain in the low AIQ category. In other words, access to AI does not equal readiness to use it well.

You can also see the AI learning problem in abandonment behavior. Gownder described employees who keep hearing they should use AI, but each attempt comes with “a really high hurdle.” They are unsure whether it is the right moment, they try, they fail, and then they stop trying. That is not just a training problem. It is an employee learning problem shaped by real workflow friction.

Why enterprise AI is a learning problem, not just a training problem

Once you define the gap correctly, the next step is to reframe where it shows up. Enterprise AI fails less often in the classroom than in the workflow.

Gownder made that explicit in the webinar: “People need to learn new behaviors in context during workflows.” That statement changes the conversation from instruction to application. Employees do not succeed with AI because they once attended a course. They succeed when they can recognize the right moment to use AI, apply it in the right application, and judge the output well enough to continue the task.

The source material shows why this is harder than many leaders expected. Gownder said early assumptions treated AI as if it would be easy because it works through natural language, but “the truth of the matter is it is not an automatic at all.” He added that AI is “the first widely used technology that makes mistakes often,” which means employees need judgment, not just access. They must know when to question outputs, when to trust them, and when not to use AI at all because of privacy, ethics, or compliance concerns.

The readiness data supports that view. While 88% of organizations are already “in the game with workforce AI,” only 22% of employees are in the high AIQ segment. Gownder also noted that 45% of information workers do not know when to question AI output, and only 36% say they know what prompt engineering is and how to use it. If your employees cannot apply AI at the right step with the right context, adoption becomes unstable. That is why the AI learning problem is really a behavior and change management problem.

What causes the AI learning problem in the workplace?

If AI adoption is a learning challenge, you need to look at the structures around the employee, not just the employee. Low adoption often reflects workflow design, weak reinforcement, and poor in-the-moment support more than simple resistance.

The first cause is lack of awareness in the flow of work. Hila Segal said one of the biggest hurdles WalkMe sees is “basic awareness,” meaning employees do not always know which tools are available or how to match the right AI capability to the right task. Gownder agreed and said the problem is not only whether employees know the tools exist. It is whether they are aware “at any given moment” that a tool could help and what it would take to create value from it.

The second cause is high workflow friction. Gownder described “abandonment behavior” where employees try AI, hit a hurdle, and give up. He also pointed to the “blank page problem,” where organizations roll out AI and say, in effect, you can do anything with it. For many employees, that creates “a moment of cognitive uncertainty.” They do not know where to start, what task AI fits, or whether the time investment will pay off.

The third cause is underinvestment in learning and support. Gownder said employers often assume people will “learn it on your own time” and called that “not a winning strategy.” He also noted that leaders consistently overestimate how many employees have completed formal training. In his words, many leaders are “gobsmacked” to discover how little attainment they actually have.

The fourth cause is growing complexity. The environment is not getting simpler. Gownder said employees now face “all these different agents” and must decide when to use them, whether to trust them, and how to collaborate with them. That complexity helps explain another important finding from the webinar: 59% of information workers are not open to collaborating with agents that conduct work on their behalf.

Taken together, those barriers describe a workplace learning problem, not an employee motivation problem alone. As Segal said in closing, the answer is not buying more tools or “asking them to become prompt engineers overnight.”

How do you solve the AI learning problem?

Once the problem is clear, the solution becomes more concrete. You solve the AI learning problem by moving from one-time instruction to in-workflow learning, behavior reinforcement, and measurable adoption.

The webinar points to three operational priorities. First, you need to meet employees where they work. Segal’s closing takeaway was direct: solving the AI adoption gap “really is about meeting people where they actually work.” That means surfacing help inside the live workflow, not hoping employees remember training from weeks earlier.

Second, you need to reduce cognitive burden at the moment of action. Gownder described low prompt and no prompt AI tools as systems that are “anticipatory” and “proactively identifying what a user needs, rather than waiting to be told or asked.” He argued that these tools can lower both the cognitive burden and the skills burden by guiding people contextually and helping them focus on work judgment rather than tool mechanics.

Third, you need a learning model you can measure. Gownder’s AIQ framework exists because organizations need a diagnostic, not assumptions. If only 22% of employees are in the high AIQ segment, you need to know where the gaps are across understanding, hard skills, soft skills, and ethics. You also need to know whether employees are actually using AI in live workflows or abandoning it at points of friction.

That is why the enterprise response must combine change management with in-workflow support. Training still matters. But on its own, it does not solve the AI learning problem.

A practical framework for enterprise AI learning

That leads to a more useful question: what should leaders actually do? A practical enterprise framework starts by treating AI learning as an ongoing operating model rather than a launch event.

Begin with assessment. Gownder said AIQ is meant to measure how ready employees are to use AI “productively and safely.” The framework evaluates understanding, hard skills, soft skills, and ethics across 20 questions. Before you scale AI further, you need a clear picture of where your teams are ready, where they need support, and where risk is highest.

Next, connect AI to specific moments in work. Segal emphasized the need to “match the right AI capability to the right task,” while Gownder warned against the blank page problem. Instead of handing employees a general-purpose tool and broad expectations, identify the workflows where AI should help and define the cues that tell an employee when to use it.

Then, reinforce learning in context. Gownder recommended more than online courses. He called for “a much more social learning environment,” where teams share tips, experiment together, and build confidence over time. He also recommended investing in low prompt and no prompt tools that guide employees in the moment.

Finally, measure adoption as performance, not as access. Leaders often know who has a license. That is not enough. You need to know whether employees are using AI in the right workflows, whether they complete tasks successfully, and where they abandon the process. That is how employee learning becomes an operational discipline instead of a hopeful assumption.

Examples of the AI learning problem in real enterprise workflows

The AI learning problem becomes easier to spot when you look at everyday work. The same pattern appears across functions, even when the tools and tasks differ.

In sales, an employee may have access to AI tools but still hesitate during a live workflow. They may not know whether this is the right moment to use AI, what tool fits the task, or how much they should trust the result. Segal said WalkMe customers are designing experiences that proactively surface the right tool “in sales or in HR or in supply chain,” because waiting for employees to figure it out themselves creates friction and abandonment.

In HR, the problem often shows up as uncertainty and confidence gaps. Gownder said more than half of employees are not confident about adapting to AI-based systems for work. That matters in people-heavy workflows where judgment, privacy, and policy all matter. Employees may avoid AI not because they reject it, but because they are unsure what is appropriate.

In operations and supply chain, complexity compounds the issue. Gownder said employees now have to decide whether to invoke an agent for a particular task, whether they trust it, and where the human needs to remain in the loop. That is a new way of working, not just a new tool.

Across all three examples, the pattern is the same. Employees do not need more abstract AI messaging. They need contextual support, clearer signals, and repeated in-workflow learning that helps them act with confidence.

What leaders should do next about the AI learning problem

The AI learning problem is the adoption gap between access and performance. Your employees may have licenses, but that does not mean they have the habits, confidence, and workflow judgment needed to use AI well.

Three takeaways stand out. Training is necessary, but insufficient. Employee learning must happen inside the workflow, where people decide whether to use AI, how to use it, and whether to trust the output. And leaders need task-level proof, not license counts, to know whether AI adoption is actually improving work.

As Hila Segal said, solving the gap is about “meeting people where they actually work” and removing friction. If proving AI ROI is the next conversation you are having with your board, the WalkMe action bar is where that proof starts. Screen-level context intelligence helps surface the right AI help in the moment, while workflow-level visibility shows whether that help is turning AI potential into AI performance.