AI is erasing the practice layer of everyday work, and L&D must rebuild it. Dmitry Zaytsev says draft writing, rough analysis and early recommendations were safe spaces for learning judgement. As machines take those tasks, organisations need structured practice, decision reviews and feedback in the flow of work again, deliberately.
Artificial intelligence is changing the way people work, but it is also changing the way people learn at work. For learning and development teams, the next challenge may not be producing more content, but rebuilding the practice that work itself used to provide.
Most conversations about AI in workplace learning focus on efficiency. AI can help create content faster, personalise learning journeys, summarise information, support coaching, and help employees access knowledge more easily. These developments matter, and learning and development teams should use them where they genuinely improve the learning experience.
AI is starting to remove some of the ordinary work people used to learn through
However, there is another side of the AI shift that deserves more attention. AI is starting to remove some of the ordinary work people used to learn through. First drafts, research passes, rough analysis, meeting summaries, early recommendations, basic comparisons and simple planning tasks were never only administrative work. They were part of the practice layer of work. They gave people a low-risk space to try, think, make imperfect decisions, receive feedback and gradually build professional judgement. That layer is now becoming thinner.
Efficient for productivity, but not for learning
In many organisations, AI can already produce the first version of a document, the first summary of a discussion, the first outline of a plan or the first answer to a problem. This can save time, but it also changes what early-stage employees and developing managers actually get to practise. If the machine produces the rough version, the human may move straight to reviewing, editing or approving. That sounds efficient, but it assumes the person already knows what good looks like.
Many people learn what good looks like by making the rough version themselves. This matters because work is one of the strongest learning environments an organisation has. People do not develop capability only through courses or formal programmes. They learn by preparing badly written drafts and understanding why they were weak. They learn by researching too broadly and then being shown how to narrow the question. They learn by making a recommendation that is almost right, then hearing a more experienced colleague explain what context was missing. These moments can feel inefficient, but they build judgement.
Human plus agentic AI?
The World Economic Forum Future of Jobs Report 2025 describes technological change as one of the major forces reshaping jobs and skills by 2030. Microsoft’s 2025 Work Trend Index also points towards a new organisational model built around hybrid teams of humans and AI agents, with systems that are AI operated but human led. This direction makes sense, but human led systems require humans who have had enough practice to lead well.
That is where the L&D challenge begins. The future employee will not only need to know how to use AI tools, but also when to trust an output, when to challenge it, when to add context, when to escalate and when to keep the decision fully human. These are judgement skills, and judgement is difficult to build through passive content alone.
A person can watch a video about critical thinking and still struggle to challenge a confident AI generated answer. A manager can complete a course on feedback and still fail to notice that an AI generated performance summary has missed the most important human context. A new employee can learn prompt techniques and still lack the professional instinct to know whether the answer is safe, fair or useful. The risk is that organisations mistake AI fluency for capability.
Application of judgement
AI fluency matters, but it is only one part of readiness. The deeper capability is the ability to apply judgement in real situations. That requires practice, feedback and exposure to the messy conditions of work. It requires people to make decisions before the answer is obvious. It requires them to see consequences, not just consume guidance.
This is especially important for early career employees. Junior work has often been treated as low value because it is repetitive, basic or easy to delegate. But much of that work has developmental value. A junior employee who drafts a client note is learning tone, structure and commercial judgement. A junior analyst who prepares the first research summary is learning how to separate signal from noise. A new manager who writes their own first feedback notes is learning how to balance clarity, fairness and responsibility.
If AI removes too much of this practice too quickly, organisations may create a future capability gap. People may appear productive earlier because they can produce better outputs with AI support, but they may not have built the reasoning beneath those outputs.
The AI-learning task gap
For L&D teams, this changes the question. The question is not simply how to use AI to deliver learning faster. The better question is where practice still happens when AI changes the work itself.
This means L&D needs to work more closely with business leaders, managers and HR to identify which tasks are not only operational, but developmental. Some tasks may be slow and imperfect, yet still essential for learning. Before automating them completely, organisations should ask what capability the task used to build.
If a task taught judgement, context, communication, prioritisation or decision making, then removing it creates a learning gap. That gap needs to be replaced deliberately.
How L&D can support
One answer is structured practice. Instead of assuming people will naturally learn through exposure, organisations can design opportunities for employees to practise realistic work decisions in safer conditions. This may include scenario-based learning, simulations, supervised case work, role-based challenges, decision reviews and structured reflection after AI assisted tasks.
For example, rather than asking a new employee only to edit an AI generated report, an organisation could ask them to compare three possible outputs, explain which one they trust most and identify what context is missing. Rather than giving managers an AI generated feedback draft and asking them to approve it, L&D could build exercises where managers decide what must be changed, what must be removed and what must be said directly by a human.
Another answer is supervised decision making. People need chances to make small calls before they are expected to make large ones. AI may reduce some of the work involved in preparing options, but employees still need practice choosing between them. Managers should be trained to review reasoning, not only outputs. They should ask why someone accepted a recommendation, what alternatives they considered and what risk they noticed.
L&D in the flow
This is where learning becomes part of workflow design. L&D cannot sit only at the end of the process, creating courses after the work has already changed. It needs to help organisations redesign how people build capability inside AI assisted environments.
The LinkedIn Workplace Learning Report 2025 argues that career development and AI adoption are becoming connected strategies for organisational agility. That is a useful direction, but career development will need to include more than access to learning content. It must include opportunities to practise the new judgement demands created by AI.
The strongest L&D teams will be the ones that protect learning inside the work, not outside of it. They will help leaders understand which forms of practice are worth preserving. They will identify where employees are losing exposure to decision making. They will build learning experiences that let people use AI, question AI and develop the human capabilities that AI cannot replace.
This also means accepting that efficiency is not always the only goal. In some cases, it may be better for a person to do the first draft themselves, even if AI could do it faster. In other cases, AI can produce the first version, but the learning task should be to critique it properly. The important point is that organisations should make this choice consciously. If every rough task disappears, people may lose the route through which they become capable of handling complex work later.
AI takeover
AI can make learning more accessible, personalised and responsive. It can support L&D teams in useful ways. But if organisations are not careful, AI can also remove the quiet practice that used to turn beginners into professionals and professionals into leaders.
The next phase of workplace learning should focus less on producing more learning content and more on rebuilding opportunities to think, apply, decide and receive feedback. In an AI assisted workplace, capability will not come from knowing the tool alone. It will come from practising the judgement needed to use it responsibly.
Dmitry Zaytsev is Founder and CEO of Dandelion Civilization
