A Remote Job Is Also a Learning System
The subject here is remote and hybrid work as a learning system inside a distributed team. The specific question is how a newcomer moves from visible work to usable judgment when colleagues are not physically nearby. Tracing that chain helps employers and workers distinguish a missing feedback loop from a general argument about office culture, and it shows why the same arrangement can feel harmless to an experienced specialist yet costly at the beginning of a career.
Learning happens in the gaps between assignments
A new employee rarely learns a job by absorbing a complete manual and then executing it alone. They learn while trying to do real work. A first attempt reveals what they understand, but it also reveals what they have not yet learned: which detail matters, which shortcut is safe, which audience needs a different explanation, and which mistake will become expensive if it travels downstream. Someone with experience can often see those differences quickly. A beginner usually cannot.
The learning sequence is therefore more granular than a training course. A newcomer produces a visible piece of work. A colleague or manager notices it, asks why a choice was made, points to a missing consideration, or demonstrates a better way to approach the problem. The newcomer revises the work and tries the judgment again. With enough cycles, a standard that was once external—something another person had to explain—becomes part of the newcomer’s own decision-making. Skill is not only information retained in the head. It is the ability to recognize a situation, choose an appropriate response, and notice when the response is failing.
That sequence can happen in a formal review, but it often starts earlier and in smaller moments. A colleague glances at a draft before a meeting, hears a question while working on the same project, or notices that a handoff is about to create confusion. The correction may take a minute. Its value comes from arriving while the work is still being formed, when the learner can connect the explanation to a concrete decision. A scheduled lesson can explain a principle; a timely correction shows where that principle changes what someone actually does.
The New York Fed’s firm-level evidence is consistent with this account. In proprietary data from a Fortune 500 company, employees who worked next to colleagues received more feedback on their output and more mentorship. When colleagues were separated even by a short distance, feedback tapered off dramatically, and the reduction was more pronounced for younger workers. The relevant difference was not simply that one group had an office and another had a home. It was that one arrangement made it easier for other people to observe work and respond to it as it happened.
The distinction between feedback and evaluation matters. An annual or semiannual performance review tells an employee how the organization has judged a period of work. It may be carefully constructed, but it arrives after many decisions have already been made. Informal feedback operates closer to the point of action. It can prevent a wrong assumption from becoming a habit, or make a tacit standard visible before a beginner mistakes it for personal preference. Mentorship adds a longer horizon: it helps a newcomer understand not only whether a task was acceptable, but how experienced people decide what deserves attention in the first place.
A beginner’s dependence on these interactions is not evidence of inferior ability. It is a consequence of having fewer reference points. Experienced workers carry examples of past problems, internal maps of who knows what, and a sense of which ambiguities are normal. A newcomer is still building those maps. When feedback becomes less frequent, the experienced worker may compensate through memory and inference; the beginner may simply proceed with an uncorrected model. The resulting gap can look like a difference in motivation or talent when it is partly a difference in access to information.
This is the difference between an instruction and a standard. An instruction says what a task requires. A standard explains what “good enough” looks like when the instructions leave room for judgment. A new employee may know how to prepare a report yet not know which uncertainty must be surfaced, how much evidence a decision needs, or when a technically correct answer will confuse the person who has to use it. Those standards are often embedded in examples and corrections rather than written down in full. Each small interaction narrows the range of possible mistakes and gives the learner a better model of the work.
The learning loop also compounds. Early feedback changes the next attempt; the next attempt gives a colleague better evidence about what the learner understands; that evidence makes later feedback more specific. If the first attempts stay unseen until a final review, the loop starts late and the correction has to cover too many decisions at once. A remote team may still be capable of delivering the final assignment, but it may be less capable of seeing the path by which a beginner arrived there. For established workers, that hidden path may not matter. For a newcomer, it is where most of the development occurs.
A simple way to see the mechanism is to follow the same piece of work through a team:
The sequence does not require every correction to happen face to face. A shared document, a recorded explanation, a code review, or a short video call can carry useful guidance. The harder question is whether the team reliably creates the conditions for those exchanges. Is the work visible before the final deadline? Does someone own the responsibility to look at it? Can a beginner ask a small question without turning it into a formal meeting? Does the person giving feedback have enough context to explain the judgment rather than only mark an error?
These conditions are easy to overlook because organizations often count the visible parts of work: meetings held, messages sent, tasks closed, or hours logged. They do not always count the unplanned observations that prevented rework or helped a new employee understand a standard. Distance can leave the formal structure intact while reducing the number of opportunities in which a learner’s incomplete understanding becomes visible. The team may still have all its meetings and tools; it may simply discover mistakes later, after they are more difficult to correct.
When a learning gap becomes a quality and hiring problem
The proposed mechanism has consequences beyond how comfortable a newcomer feels. In the New York Fed’s firm data, employees who had previously worked side by side with teammates—and therefore had received more mentorship—produced higher-quality output when everyone later worked in isolation than employees who had spent more time working at a distance. The authors also report that teams experiencing a larger change in proximity after return-to-office mandates showed greater improvements in work quality.
That result is easy to misread. It does not demonstrate that an office independently creates quality, or that every employee who works remotely produces worse work. One plausible interpretation is cumulative: earlier proximity helped some employees acquire standards and judgment, and that accumulated learning remained useful when the work environment changed. The quality difference may therefore reveal what the team had already taught people, not just what a desk location did on a particular day.
It also matters that the evidence comes from one Fortune 500 online retailer and is especially compatible with software engineering, code review, and other work with visible outputs. The source does not establish the same relationship for every knowledge-work occupation. A financial analyst, a consultant, and a healthcare professional may have different forms of tacit knowledge, different rules for handling confidential work, and different ways to make progress observable. The portable claim is about a mechanism—feedback and mentorship can weaken when distance removes low-friction observation—not about a universal office effect.
The mechanism becomes economically important when a firm decides whom it can afford to hire. The New York Fed reports that while the studied company’s offices were closed during the pandemic, it hired fewer inexperienced workers and more experienced workers. After the offices reopened, hiring shifted back toward younger workers. The contrast persisted within the organization: positions on distributed teams continued to draw more experienced hires even after reopening, while the firm was more willing to hire younger workers when proximity to colleagues was feasible.
The authors interpret this pattern as a training constraint. An experienced employee can begin with a larger store of relevant judgment, so the team may spend less time explaining basic decisions or correcting early attempts. A beginner may be a good long-term investment, but the investment is harder to make when the team cannot easily see what they are trying, cannot answer small questions at the moment they arise, or cannot give enough context-rich feedback without scheduling additional work. If a firm’s immediate priority is reliable execution, hiring experience can appear to be the safer choice even when it weakens the supply of people who will become experienced later.
This is a decision under uncertainty, not a simple ranking of workers. Before hiring, a manager cannot observe how quickly a graduate would learn if given good guidance. They can observe credentials, prior experience, and the apparent amount of supervision a role will require. When the cost of teaching is easy to absorb, the firm can take a chance on potential and let real work reveal the person’s strengths. When teaching is difficult to coordinate, prior experience becomes a convenient signal for readiness. The result can be self-reinforcing: the people who already had a chance to learn look safest, while the people who need a first chance appear expensive because the system has not made learning cheap.
That is why the feedback mechanism matters even when no one intends to exclude beginners. A distributed team may not make a conscious decision against entry-level hiring. It may simply discover that deadlines, handoffs, and manager attention leave too little room for unplanned teaching. Hiring then shifts toward workers who can operate with less support. The aggregate effect can be a narrower entry route into a field even though every individual hiring choice looks defensible.
That interpretation should remain conditional. The hiring pattern comes from one firm and is descriptive rather than a controlled test of managerial motives. Worker availability, team composition, changing business needs, and other pandemic effects could have influenced the choices. “The company hired more experienced people because remote training was harder” is stronger than the evidence allows. A more defensible statement is that the observed hiring behavior is consistent with a training-cost explanation, and that the explanation fits the feedback and mentorship pattern found in the same firm.
This is also where natural proximity and designed remote work part ways. Physical closeness does not guarantee good teaching. An office can contain distracted managers, unclear standards, and juniors who are afraid to ask for help. Nor does distance make strong apprenticeship impossible. A distributed team can make work visible early, assign a named mentor, schedule review around real attempts, and reserve synchronous time for ambiguity rather than status reporting. Those choices replace some accidental encounters with deliberate ones.
The replacement has to be reliable, not merely available. A document repository is not mentorship if nobody reads a beginner’s draft. An open calendar is not access if a newcomer cannot tell which questions are worth asking. A group chat is not feedback if responses arrive after the decision has already propagated. The test is whether the team supplies a repeated loop of attempt, observation, explanation, revision, and increasing independence.
For established employees, that loop may be needed only at the edges of unfamiliar work. For newcomers, it is the work. This is why the same remote arrangement can have different consequences by seniority without implying that junior workers are inherently less productive. Their task is not only to deliver an output; it is to acquire the standards that make future output dependable. If those standards remain invisible, the organization may not notice the learning deficit until quality falls, a promising employee leaves, or hiring managers quietly stop taking chances on beginners.
The next question is therefore empirical rather than symbolic: when an organization actually tests a moderate hybrid schedule, what happens to the performance and retention of employees who already belong to it? That comparison can clarify what flexibility protects—and, because of who was included or excluded, what it still leaves unresolved about the first rung.