How Generative AI Is Changing the Role of L&D in Business

About the Research

Generative AI is already changing the day-to-day work of L&D.

But based on the interviews we conducted, the most interesting change is not simply that AI makes it faster to create a course or a test.

The deeper change is in how the task itself is defined.

If learning materials can be prepared in hours instead of days, the main constraint gradually shifts away from content production toward understanding what content is actually needed, who needs it, and what change in their work it is supposed to create.

This research explores that shift through the experience of eight professionals working in different organizational contexts.

We did not set out to produce a statistically representative picture of the market. We were interested in the practical observations of people already working with L&D, learning technologies, organizational change, and AI.

The result is a set of several relatively consistent observations — along with several more ambitious hypotheses that cannot yet be considered established findings.

Methodological Overview

The report is based on 8 semi-structured in-depth interviews, each lasting between 60 and 90 minutes.

Participants have experience in:

• industrial companies;
• financial and technology sectors;
• mining;
• e-learning and LMS;
• international business;
• AI technologies;
• small and medium-sized businesses.

The final version of the research uses anonymized data. Names and company names are included only to the extent agreed upon with the participants.

Participants

How to Read the Findings

We grouped the observations according to the level of support within the sample.

Consistent pattern — 6–8 interviews

The topic appeared repeatedly across different organizational contexts.

These are the strongest observations in the study, but even they should not be interpreted as statistically proven market-wide patterns.

Recurring tendency — 3–5 interviews

The topic appeared across several interviews, but may depend on industry, company size, or level of technological maturity.

Hypothesis or scenario — 1–2 interviews

An interesting observation that is not yet sufficiently supported by the sample.

This distinction matters: we do not want to turn the ideas of one strong expert into the “voice of the market.”

1. AI Is Already Significantly Accelerating Routine L&D Work

Consistent pattern

Support: 8 of 8 interviews

All participants described, in one way or another, a significant acceleration in working with text, structuring information, preparing materials, or handling other tasks that previously required substantial manual effort.

The most frequently mentioned areas included:

• preparing drafts;
• creating content structures;
• developing test questions;
• preparing technical specifications;
• reworking existing materials;
• adapting content;
• working with large volumes of information.

At the same time, estimates of the actual acceleration vary considerably depending on the task. Some interviews included estimates of several times faster, but we do not treat these as a universal productivity coefficient.

One participant described the change this way:

“I used to spend a week writing and struggling with a technical specification for developers... Now it takes a couple of hours and is formulated beautifully. Sometimes I even have to remove excessive requirements.”

— Respondent 6

The important point is not the specific acceleration figure, but the nature of the change: AI is moving a significant share of the work from manual production toward task definition, editing, and quality control.

2. As Content Becomes Cheaper, Errors Become Cheaper to Produce Too

Consistent pattern

Support: 6 of 8 interviews

Several participants independently described the downside of mass generation.

When creating materials becomes easy, an organization gains the ability to create more content than it can effectively review and use.

Possible consequences include:

• local teams independently creating inconsistent materials;
• the same content appearing in multiple versions;
• employees receiving excessive amounts of information;
• persuasive AI-generated text remaining unchecked;
• the volume of available content exceeding the organization's ability to select what matters.

This changes the role of L&D.

If a significant part of the work used to be production, the importance of editing, verification, curation, and contextualization now increases.

In other words, AI can make content cheaper to create, but it does not eliminate the need to answer the question:

“Do we need this content at all?”

3. Diagnosing the Real Request Remains a Human Domain

Consistent pattern

Support: 7 of 8 interviews

One of the most consistent themes in the research was the need to understand the actual cause of a business request before designing learning.

Participants described situations in which the business arrives with a ready-made solution:

“We need a course.”

But the underlying problem turns out to be related not to a lack of knowledge, but to a process, incentive system, tools, management, or work organization.

One participant put it this way:

“What remains resistant to change is identifying the problem and unpacking the request. When a client comes and says, ‘Train us on this,’ — figuring out the real cause and whether a course is needed at all is something AI cannot do.”

— Respondent 8

This does not mean AI cannot help with diagnosis.

It can analyze data, group information, identify patterns, and propose hypotheses.

But responsibility for defining the problem remains with people.

And this may be one of the most important shifts for L&D: as content production becomes cheaper, the value of defining the right task increases.

4. L&D May Be Shifting from Production to Design

The previous observations point toward a broader conclusion.

If AI takes over an increasing share of learning-material production, L&D may gradually shift its efforts toward several other areas:

This is not yet a single new model of L&D. Rather, it is a direction of change that appeared across several interviews.

5. AI Is Gradually Bringing Learning Closer to the Workflow

Another theme across the interviews concerns where learning actually happens.

The traditional model assumes a separate learning environment:

course → learning → assessment → work.

AI makes another sequence increasingly possible:

task → assistance → action → feedback.

For example, an employee may not need a separate course on how to use a system if, at the moment of performing a task, they can receive a hint, explanation, or example directly inside the work tool.

This does not mean courses will disappear.

For foundational knowledge, certification, complex programs, and many other purposes, dedicated learning will remain relevant.

But the boundary between learning and work may become less distinct.

6. “Hollow Skill”: When the Result Exists but the Skill Does Not

Recurring tendency

Support: 3 of 8 interviews

Three participants independently described a related problem: AI can allow a person to successfully complete a task without actually mastering the underlying skill.

One participant called this a “Hollow Skill.”

“The Hollow Skill effect means that you perform a task with AI and consider yourself to possess that skill. But if AI is removed, the skill isn't there. We are taking cognitive credit from our future: by using quick solutions, we become less capable in the future.”

— Respondent 5

Another participant described a similar effect in more practical terms:

“AI takes away the moment of thinking: they hand everything over to it, and afterwards they can't even put two words together.”

— Respondent 7

This is not enough to claim a systemic degradation of employee skills.

But it raises an important practical question for L&D:

What should a person be able to do independently, and what can they perform together with AI?

This distinction becomes especially important in onboarding, certification, and professions where mistakes are costly for the organization.

7. Human Interaction Does Not Automatically Disappear

Recurring tendency

Support: 5 of 8 interviews

Several participants noted that digital habits do not automatically mean a preference for AI in every learning format.

This was particularly visible in discussions about younger employees, mentoring, and complex social situations.

One participant gave a concrete example:

“I brought together a group of 150 interns and asked what they would choose: a live two-day intensive or a 90-minute microlearning course. Everyone raised their hands for the live instructor.”

— Respondent 3

This example is interesting, but it cannot automatically be turned into a conclusion about “Gen Z” as a whole.

Within our research, it is better understood as evidence that the need for live interaction may persist even among audiences that actively use digital tools.

AI and human interaction do not necessarily represent interchangeable formats.

8. The Role of the Mentor May Change — But Not Necessarily Disappear

Hypothesis

Support: 2 of 8 interviews

The interviews also produced a more radical vision of the future.

If a personal AI has access to a person's work context, it could potentially support them continuously: explaining, suggesting, finding knowledge, identifying gaps, and helping solve tasks.

One participant compared this to having a “second brain”:

“Generals used to have orderlies. Now neural networks allow every soldier to have one. A second brain appears, one that knows your context and highlights your weak areas...”

— Respondent 2

Another participant, by contrast, saw an AI mentor as a potential threat to the institution of mentoring itself:

“Creating an AI mentor on your phone is really cool. But once you create it, you essentially kill the live institution of mentoring and knowledge transfer inside the company.”

— Respondent 1

This illustrates why future scenarios cannot be reduced to a single trajectory.

The same technological shift can simultaneously increase access to support and challenge existing forms of human interaction.

9. L&D Faces Not Only a Technological Problem, but an Institutional One

Contextual tendency

Support: 4 of 8 interviews

Several participants critically described the existing system for evaluating learning.

In particular, they pointed to the focus on participant satisfaction and metrics such as NPS, which can be relatively easy to improve through design, service, and the emotional impact of a program.

One participant described the problem quite bluntly:

“We measure NPS and satisfaction, pushing scores up to 9.7 out of 10 through design and wow effects. KPIs turn into salaries and bonuses. But the business does not change.”

— Respondent 6

It is important not to turn one expert's position into a claim that NPS is useless.

A more cautious conclusion is this:

satisfaction with learning and change in behavior are different outcomes, and the first does not guarantee the second.

For L&D, this means looking for additional ways to measure outcomes: skill application, quality of work, speed of adaptation, behavioral change, and other indicators connected to a specific business task.

10. AI Economics: Generating Content “On the Fly” Is Not Always Rational

Hypothesis based on experience with large-scale systems

Support: 2 of 8 interviews

A separate theme emerged when participants discussed large-scale learning systems.

One participant described mass generation of educational content for hundreds of thousands of employees and argued that continuously generating large volumes of content for every user can create significant infrastructure costs.

“A fire-safety course for 300,000 people... if you multiply the number of characters by the number of people, the training costs tens of millions of rubles [in tokens]. There is no upside. Generating and checking it in advance is several times cheaper.”

— Respondent 4

This does not prove that dynamic generation is generally economically inefficient.

Rather, the interviews suggest that it is useful to distinguish between two approaches:

generate once and reuse many times

and

generate a personalized result every time someone asks.

The optimal approach depends on audience size, task complexity, AI model costs, and the required degree of personalization.

11. The Most Radical Scenario — Zero L&D

Scenario hypothesis

Support: 1 of 8 interviews

One participant proposed the most radical perspective: if personal AI becomes a permanent assistant to employees, a significant part of the corporate university's functions could become unnecessary.

In this model, employees would acquire knowledge not through a centralized learning catalog, but directly through their own AI assistant, which understands their work context.

This can be provisionally called the Zero L&D scenario.

Importantly, this is not a finding of the research and not a forecast.

It is one participant's hypothesis, interesting precisely because of its radical nature.

It raises the question:

If AI can genuinely provide employees with personalized development directly within their workflow, which parts of the traditional L&D function will still be necessary?

12. What Is Supported — and What Remains a Hypothesis

This table may be more important than any polished final formula.

It shows where we see a recurring signal and where we still have only an interesting idea.

13. What Does This Mean for L&D?

The research does not provide a ready-made model for the future L&D function.

But it allows us to identify several directions for further discussion.

1. Shift attention from production to problem definition

If the first version of content can be created in minutes, value shifts toward understanding the problem.

L&D needs not only to know how to create learning, but also how to ask:

“Is learning actually needed here?”

2. Make verification a distinct capability

AI-generated content requires editorial and subject-matter review.

This applies not only to factual accuracy, but also to fit with the actual work context.

3. Distinguish AI assistance from mastered skill

For some tasks, employees may simply need to know how to work effectively with AI.

For others, they need to possess the underlying skill independently.

These two requirements should not be conflated.

4. Do not frame AI and humans as opposites

Automation does not necessarily have to replace live learning.

AI may instead free up human time for formats where human interaction genuinely matters.

5. Rethink what L&D measures

The number of courses, learning hours, and participant satisfaction may remain useful operational indicators.

But they do not answer the most important question:

Did anything change in the person's work after the learning?

14. What Does This Mean for Business?

For business, the main conclusion of the research can be stated quite simply:

AI should not be evaluated in L&D only by how much cheaper it makes content production.

A more important question is what the organization gets as a result.

If AI allows an organization to create five times as many courses, but employees do not perform their work better, the economic effect may be close to zero.

If AI allows the organization to diagnose problems faster, give employees help at the right moment, and free L&D specialists to focus on more complex tasks, the effect may be very different.

AI adoption therefore makes sense not simply as a content-production automation project, but as an opportunity to rethink the very architecture of employee development.

15. Conclusion

Eight interviews do not provide grounds for claiming that L&D will disappear.

They show something else.

AI makes learning production cheaper and faster. And precisely because of this, it begins to question the need to produce so much learning in the first place.

When a course, test, or instruction is no longer a scarce resource, value moves elsewhere:

understand the problem → choose what is needed → embed support into the work → verify the result → preserve human interaction where it genuinely matters.

This may lead to a change in the role of L&D — from a function that primarily produces and administers learning to a function that designs the conditions for development and change in work.

But the eight interviews do not yet allow us to say how far this transition will go.

And that may be the most useful result of the research.

We have not obtained a final answer about the future of L&D. We have obtained several questions that can now be tested on a larger sample.

From training to development — the transition is not complete. But AI is clearly accelerating movement in this direction.

Appendix. Anonymized Participant Profiles

Respondent 1 — Head of Corporate Learning Methodology, industrial sector.

Respondent 2 — Organizational change and AI specialist, financial and technology sector.

Respondent 3 — Head of Assessment and Development, mining industry.

Respondent 4 — Educational product and AI expert, financial and technology sector.

Respondent 5 — AI competency and AI architecture expert.

Respondent 6 — E-learning and LMS expert, 15 years of experience.

Respondent 7 — Head of Learning at a large retail company.

Respondent 8 — Head of L&D at a small company.