How do you gain experience when AI does more of the work?
In a training session, participants are given a difficult architectural problem. They try to solve it, discuss the consequences of their decisions, and then the trainer shows them a solution that looks obvious in retrospect. Everyone nods. Yet they would not have known how to get there themselves.
Sławomir Sobótka, founder of Bottega IT Minds, describes this as pulling a rabbit out of a hat. His team set itself a harder task: break down how experts work into steps that others can learn. Today the question returns in a new form. If AI can produce an answer in seconds, might it also remove an opportunity to understand the problem?
An answer is the start of the work
Sobótka speaks mainly about building complex software, the work his company does. In such a system, a decision about code can affect other parts of the product, its performance and the ease of changing it later. An AI response can look convincing while introducing problems a beginner does not yet know how to recognise.
In his view, an expert understands the consequences of decisions and has more than one way to act. The question is not simply whether a solution works today. It is also when it might stop being sufficient and what changing it would cost.
That is why years of service can be misleading. Someone may spend twenty years applying the same set of tools to similar problems. They become practised, but have few opportunities to compare different approaches.
From instructions to judgement
When training teams, Sobótka draws on the Dreyfus model of skill acquisition. He explains it through learning to drive. A novice needs instructions: when to turn the wheel and how to perform a manoeuvre. With time, the learner can choose the steps, then recognise recurring situations and correct their own mistakes.
Further development calls for a wider view. A proficient person starts asking why a familiar rule works in one situation and fails in another. An expert recognises more aspects of a problem at once, although they may not immediately be able to explain the path that led to an answer.
Sobótka and his colleagues tried to narrow that gap in their training. Participants first grapple with a task, then examine the consequences and learn concepts that help them name what happened. The team also worked on methods for moving from a problem to a solution, rather than merely showing the finished result.
Faster work does not necessarily mean more learning
Sobótka does not have a definitive answer to how widespread use of language models will affect the development of expertise. He cautions that the tools are relatively new and says he does not know whether there is enough reliable research on their effect on that process.
In the teams he works with, however, he sees AI accelerating people who already have a foundation. That acceleration can extend to mistakes. Someone who does not understand the problem or know how to assess the result may get an answer faster without noticing its weaknesses.
The question for beginners remains open. When a tool takes over tasks once done by junior staff, organisations need to consider where younger workers will gain practice. Sobótka observes that some companies still invest in developing newcomers, while many teams expect people to become productive quickly. He offers no single prescription for the whole industry.
Seeing the consequences of decisions
Experience requires varied situations and time to examine what follows from them. Sobótka describes a tension: someone who changes projects quickly sees many environments but may miss the long term effects of their decisions. Someone who stays in one place can observe those effects but may have little to compare them with.
His consultants enter different companies, often when a system built over many years has become difficult to extend. They can see both solutions that worked in one context and limitations that emerged years later. Not everyone has access to that perspective, but it illustrates what gets lost when experience is measured only in years.
Writing and focus when working with AI
The conversation also returns to the ability to express ideas. Working with AI agents requires describing requirements and constraints clearly enough to assess the result. Sobótka points out that writing is difficult for many people who do not practise it regularly.
Focus matters too. Running several agents in parallel can mean frequent switches between tasks. Sobótka discusses fatigue reported by some programmers and the loss of time for deep work. He asks how much effort it takes to practise concentration if we spend hours each day becoming accustomed to distraction.
At the end, he offers another way to think about development. An expert can begin learning a new field from scratch, even though that means returning to the role of a novice. He does not present this as advice for everyone. It is an opportunity to return to a position where asking questions, trying things and observing the consequences are part of the work.
This article is based on Krzysztof Kosman's conversation with Sławomir Sobótka on the EdTech Dots podcast.
