Who Teaches the Next Expert

The new employee has produced a very good answer. The answer is so good that the senior colleague asks how she reached it.

She turns the laptop around.

That settles one question. It leaves another open. Has she learned to solve this kind of problem, or has the team learned that she can obtain a solution from a tool?

Both may be valuable. A person does not need to derive every piece of knowledge from scratch before using it. We would get very little done if every journey required us to invent the wheel. But the distinction matters when this employee must check a doubtful answer, explain a decision, choose a different method, or work somewhere that uses another system.

It matters to her future earnings. It matters to the business that will eventually need experienced people. And it matters to ownership, although that connection is easy to miss.

A society can distribute small financial stakes while keeping knowledge and authority behind a narrow door. People might receive income yet remain dependent on a few institutions to tell them what is happening, whether their assets are managed well, and which choices are even available. I want a wider kind of participation than that. Ordinary people should gain capital and the practical freedom to understand, question, build, and leave.

There is a dividend in becoming more capable. We should decide who receives that one too.

The work a beginner gets to do

Think about a task you can now do almost without thinking. There was probably a stage when it was awkward. You missed something obvious. Someone corrected you, or the result corrected you. After enough attempts, you began to notice the problem before it became a problem.

Not all of that struggle was useful. Some was bad teaching. Some was needless repetition. Some was a system wasting your time and calling the waste a tradition. AI may remove a great deal of this. An explanation can be available when a colleague is busy. A person can ask a basic question without feeling exposed. A worker can compare approaches, translate unfamiliar terms, or rehearse an argument before taking it into a room.

Those are real possibilities worth pursuing. The danger comes from counting only the finished answer.

A business may find that an experienced worker with AI can produce the drafts once assigned to three beginners. It has a reason to change the workflow. But if those drafts were also how beginners learned the trade, the business has removed a route into expertise. Keeping the old task unchanged is only one possible response, and often an expensive one. The better question is what should replace its teaching function.

Some activities were doing two jobs. They produced today’s work and trained tomorrow’s worker. A cost calculation that counts the first job and forgets the second can make a saving look larger than it is.

The consequences might not appear this quarter. A team can live for a while on experience accumulated under an older system. Then a senior person leaves, a new problem arrives, or the tool produces a convincing mistake. Suddenly the missing training has a price.

This is a risk to examine, not a forecast that every AI-using firm will run out of experts. Tools can also make training better. The result will depend partly on whether someone is responsible for it and whether learning receives time in the budget.

A company should not be able to declare that everyone must adapt while quietly removing the hours in which adaptation happens.

The answer and the ability

A study of mathematics learning makes one part of this issue visible. Hamsa Bastani and colleagues tested two GPT-4-based tools with nearly a thousand students at a high school in Turkey. Both improved performance during assisted practice. On the subsequent exam without assistance, students with the more open tool did worse than the control group. The tutor designed with learning safeguards largely avoided that harm, but did not show a positive exam effect. This was a short-term school experiment, not a study of workers’ careers.¹

The distinction it asks us to notice travels further than its numerical result: doing well with help and learning to do well are different outcomes. An employer should measure the outcome it actually wants.

If the task is sending an accurate routine reminder, fast assisted completion may be enough. If the task is learning how to spot a dangerous assumption, the process deserves more attention. If a person is being prepared to take responsibility for a decision, a polished answer alone cannot establish readiness.

We can bring this back to the new employee. The senior colleague could ask her to explain the recommendation without reading it aloud. Which fact would change her mind? Which part of the evidence is weakest? What did the tool leave out? What would she do if the recommendation were challenged by the person affected?

These questions do not require an artificial ritual of doing everything the hard way. They test the ability the role will need. In some jobs, the important skill is recognizing when to call someone else. In others, it is using a trusted reference well. Independence does not mean isolation from tools or people. It means enough understanding to direct help rather than merely follow it.

That is also a form of power. A person who understands the process can ask for reasons, propose alternatives, and notice when a confident instruction serves someone else’s convenience. A person who is allowed only to press the approved button has fewer ways to disagree.

Give learning somewhere to live

Imagine a team that has saved time on ordinary reports. It could use all the saving to increase the target. Instead, it sets aside a modest block each week for practice and review. The block is paid work, visible on the schedule, and protected from routine requests. A genuine emergency can interrupt it. Ordinary impatience cannot.

The team chooses examples from problems it has actually handled, removing confidential details and using only material it has permission to use. Some examples are simple. Others contain the sorts of ambiguity that a neat training exercise tends to remove. The point is to practice judgment without making customers the test subjects.

One session might begin with a first attempt before the tool is opened. Another might compare a human draft and an AI draft. In a third, the tool supplies alternatives and the learner must explain a choice. A senior worker reviews enough of the process to see where understanding is growing and where help is needed.

There is no universal sequence here. A novice learning a language, an engineer checking a system, and a care worker organizing records need different kinds of practice. The design should follow the work. It should also change when the evidence from the team says it is failing.

The important agreement is that learning has an owner, a time, and an outcome. Someone is responsible for preparing it. The person learning is not expected to do it after an exhausting shift. Success means more than attendance at a product demonstration.

A manager may object that protected time is costly. It is. Training always consumes resources, even when the cost is hidden inside a beginner’s slower work. AI gives us a chance to make that investment more deliberate. Saving time and reinvesting some of it in capability can be a productive bargain.

The worker also has a reasonable question: capability for whose benefit?

Training that travels

A company may be willing to teach you its own system while offering little that helps you elsewhere. That can be sensible for the company and limiting for you.

You need to know the local process. You also need some knowledge that survives a login being switched off. How to define a problem. How to examine evidence. How to understand the field you work in. How to check a calculation or explain a decision. Which permissions and privacy duties apply when information moves. The exact list depends on the job, but the principle is simple: training should not make you more useful only inside one enclosure.

This belongs in negotiations about the gains from AI. A worker can ask whether the training is paid, whether it leads to a recognized qualification where one is relevant, and whether there is a record of the skills learned. A team can ask whether junior employees receive real assignments with feedback or simply monitor a stream of machine output.

None of these questions guarantees a favorable answer. People with weak bargaining power may have little room to press them alone. Collective agreements, professional standards, public training funds, and procurement conditions can help make the question larger than one employee’s courage. The legal tools vary by place. The aim is a practical one: a person’s future should not depend entirely on an employer choosing to be generous.

Ownership can widen that room as it grows. A worker with assets and some income outside the job may have more time to train or greater freedom to leave. A tiny balance will not perform miracles. An illiquid account cannot pay this month’s bills. We should judge the design by what it lets people actually do, rather than by the attractiveness of the statement they receive.

And learning cannot substitute for ownership. We should reject a bargain that says, in effect, the owners keep all the capital while everyone else gets another course. Better skills can improve your position, but they do not by themselves give you a claim on profits when a machine performs a task you once sold.

Training is a way to expand a life. It should not become an endless entrance exam for the right to have one.

The knowledge that enters the system

There is another person in this story: the experienced worker whose judgment helps make the tool useful.

She may spend hours showing which answers are wrong, revising examples, explaining exceptions, or turning years of practical experience into instructions someone else can follow. When the new system works, that labor can disappear from view. The final output has the software’s name on it. The knowledge that made it dependable has many names, some of which never appear in the presentation.

We should make that contribution visible without pretending that each answer has one owner or that every legal right is already settled. Work often draws on shared practice built over a long time. A team might reasonably use a group contribution record rather than try to attach a price to every correction.

What matters is that the introduction of AI does not become a process in which workers are asked to teach the system and then told the system proves their work had no value.

Here a standard ownership contribution becomes especially compelling. Instead of arguing over whether one prompt caused one dollar of profit, the firm can adopt a broad rule for sharing capital with the people who help create and maintain the business. Wages compensate the present work. A stake gives people participation in its longer future. The details of that stake still matter: vesting, voting, diversification, what happens on departure, and whether lower pay is being smuggled into the offer.

The standard should reach beyond the glamorous expert. The person who notices that a form confuses customers may improve the system. So may the person who finds a recurring error in a warehouse record. A business cannot discover that knowledge is distributed and then distribute the reward only to the executive who announces the discovery.

This is a moral claim and a proposed way to organize incentives. We should test its effects honestly. Employee ownership does not remove disagreements, eliminate poor management, or guarantee profits. It does change the question from whether workers might receive a discretionary thank-you to what continuing rights they hold.

A member who can ask why

Universal ownership would bring millions of people into relationships with funds, trustees, and corporate decisions they may not have had to inspect before. We should design those relationships for ordinary citizens, including people with little time or financial training.

That means plain statements. What do I own? What came in? What did it cost to manage? What can I withdraw? What is at risk? Who represents me? How do I change that representation or challenge a mistake?

A person should not need a finance degree to receive what the law says is theirs. Nor should someone be required to trade shares actively to count as a serious owner. Many people should be able to leave the day-to-day investment work to accountable professionals and get on with their lives.

But delegation needs ways to see and act. A citizen education program that explains only compound returns while avoiding voting rights, fees, conflicts of interest, and appeal procedures teaches half the subject. The missing half is often where power sits.

The same is true at work. Teaching people to operate AI is useful. Teaching them to question its deployment, understand the gain, and participate in decisions about that gain is a fuller preparation for the economy we are building.

We should expect the owners of a democracy’s productive wealth to have different ambitions. One person wants to start a business. Another wants to work fewer nights. Another wants to become excellent at a craft. Someone else wants to study a subject that will never raise their salary. Shared capital should widen that range, not require everyone to justify their choices with a forecast of higher earnings.

The new employee closes the laptop. The senior colleague asks the question again, more carefully this time. She is not being invited to recite the machine’s answer. She is being invited into the work of judging it.

A fair AI economy should offer that invitation widely. It should also offer a stake in what the judgment helps to build. We can make people more capable and more secure at the same time. That is a far better ambition than keeping them permanently busy proving they can still compete.