Organisations spend on training and treat it as the mechanism by which people get better. Most people, asked where they actually learned their job, describe something else entirely: a specific piece of work, a person who corrected them, a mistake with visible consequences.
Understanding which is which matters, because the things that actually produce capability are the things most easily lost — particularly in distributed work.
For an applied reference, see Monitask’s interpersonal synchrony overview.
The three mechanisms
Doing work slightly beyond your current ability, with a safety net. The primary mechanism. The work has to be real, it has to be hard enough to be uncertain, and the consequences of failure have to be recoverable. Take away any of the three and it stops working: work that is easy teaches nothing, work with no safety net teaches caution, and simulated work teaches simulation.
Watching someone competent, closely. How they approach an ambiguous problem, what they check first, what they ignore, when they stop. Almost none of this is articulable by the person doing it — ask an expert how they decided and you get a reconstruction, not the actual process.
This is why apprenticeship works and why it is hard to replace. It is also the mechanism most degraded by distance.
Correction with a short feedback loop. Being told, soon, and specifically, that something was wrong and why. Annual reviews do not do this. A comment on your work the same day does.
What training is actually for
Not nothing, but narrower than the budget implies.
Facts and procedures. Compliance requirements, how a specific system works, regulatory content. Training is efficient for these and there is no better alternative.
Vocabulary and frameworks. Giving someone the names for things they have half-noticed. Genuinely useful and frequently the real value of a course — not the content, but the language to think with.
Signalling and credentials. Where a certification is required or valued externally.
Permission. Sometimes the value of sending someone on a course is that it tells them they are expected to grow. That is a real function and it is worth being honest that it is the function.
What training does not do is produce judgment. Judgment comes from consequence, and a course has none.
Why distributed work makes it harder
Not fatally, and the mechanism is specific rather than general.
Observation is the casualty. You cannot watch how someone works if you cannot see them working. The incidental exposure — overhearing how a colleague handled a difficult call, seeing what someone does when they get stuck — largely disappears.
Correction gets slower and more formal. In person, a small correction is a two-second remark. Remotely it requires composing a message, which raises the threshold, so small corrections do not happen and accumulate into larger ones.
Weak ties do not form. Learning frequently comes from people outside your immediate team, and those relationships form by accident in shared space.
The problem is worst for the newest people, who have the least existing context and the fewest relationships to draw on.
None of this argues for a general return to offices. It argues for deliberately replacing what proximity used to provide — which is the actual work, and which almost nobody does. See the office as a coordination device.
What to do about it, deliberately
Work aloud. Narrate reasoning in writing where others can see it. A decision record that includes what was rejected and why teaches more than the decision alone. See documentation as infrastructure.
Pair on real work. Not a training exercise. Two people on one genuine problem, with the less experienced one driving.
Review work, not just outcomes. Looking at how something was done, with the person, is the highest-return activity a manager has and the first thing cut when busy.
Make correction cheap and frequent. Small, immediate, specific, and normal. If correction only happens in formal settings, it only happens rarely and it lands hard.
Assign work slightly beyond current ability, on purpose. With a named person available when it goes wrong. This is what development actually consists of, and it costs nothing but attention.
Protect the safety net. People learn from stretch work only if failure is recoverable. In an organisation where mistakes are expensive, people stop taking work they might fail at, and development stops.
For an individual
Ask for work you are not sure you can do. The single most effective thing available to you.
Find someone to watch. Ask to sit in, to pair, to see the working file rather than the finished output. Most people say yes and are flattered.
Ask for correction explicitly. "What would you have done differently?" gets an honest answer where "any feedback?" does not.
Write down your reasoning before you find out the answer. Then compare. It is the fastest way to calibrate judgment, and almost nobody does it.
Treat courses as vocabulary, not capability. Useful for names and frameworks; the capability comes afterwards from applying them to something real.
The uncomfortable implication
If most learning comes from stretch work with a safety net, then an organisation under sustained pressure — where everything is urgent and nothing can be allowed to go wrong — is an organisation where nobody develops.
That shows up years later as a shortage of people ready for the next level, and it is invariably attributed to hiring rather than to the conditions that stopped anyone growing. See seniority: what actually changes.
For broader public guidance and background, consult the U.S. Bureau of Labor Statistics.