Skip to main content

What Remote Work Took From Engineering Mentorship, and What Replaced It

Junior engineers used to learn by overhearing seniors debug. Distributed teams built substitutes, but only some mentorship functions survive without physical proximity.

Updated 7 min readMarkdown
What Remote Work Took From Engineering Mentorship, and What Replaced It

The desk that used to teach people things

A junior engineer at a ten-person software shop used to learn a surprising amount without anyone intending to teach them anything. They'd hear a senior engineer mutter about a flaky test while walking to the kitchen. They'd catch the tail end of someone explaining a database migration to a teammate and absorb the reasoning without being part of the conversation. They'd get pulled into a five-minute huddle at someone's monitor because they happened to be standing nearby when a bug got weird.

None of this was designed. It was a byproduct of bodies occupying the same room for eight hours a day. Call it osmosis: the transfer of judgment, debugging instinct, and institutional context that happens not through formal instruction but through proximity to people who already have it. It was never written into anyone's job description, which is exactly why it was easy to lose when teams went distributed and nobody noticed it was gone until the junior engineers started taking longer to develop good instincts.

What osmosis was actually doing

It helps to separate what in-person proximity actually provided, because "mentorship" is really several different mechanisms bundled together.

The first is exposure to reasoning under uncertainty. Watching a senior engineer debug a production issue live teaches more about how to think than reading their eventual fix. You see the dead ends, the wrong hypotheses, the moment they realize they're looking in the wrong file. None of that survives into the commit history.

The second is incidental context transfer: picking up why a system is shaped the way it is, which parts are fragile, which conventions are sacred and which are just habit. This context usually isn't written down anywhere because it lives in people's heads and gets transmitted in passing.

The third is low-stakes access. A junior engineer interrupting a senior engineer's afternoon with a half-formed question is a much smaller ask in person, where it costs thirty seconds, than it is over a scheduling link, where it costs a calendar invite and a day's wait.

The fourth is calibration: learning what "good" looks like by being around it constantly, not by reading a style guide.

Distributed teams didn't lose all four equally. Some of them translate into deliberate process reasonably well. Others don't translate at all.

The substitutes teams actually built

Most distributed engineering organizations have converged on a similar toolkit, even when they arrived at it independently.

Recorded pairing sessions try to preserve the first mechanism, reasoning under uncertainty, by capturing a senior engineer working through a problem on video. Structured onboarding buddies try to preserve the third, low-stakes access, by assigning a specific person a new hire is explicitly allowed to bother. Written design docs try to preserve the second, context transfer, by forcing architectural reasoning into a document before code gets written. Office hours, scheduled blocks where anyone can show up with a question, try to recreate some version of the fourth and third mechanisms at once.

Each of these is a real attempt to formalize something that used to be free. And each works for part of the job and fails for another part.

Where the substitutes genuinely work

Written design docs are probably the strongest substitute on this list, and in some ways an improvement on the original. A design doc forces a senior engineer to articulate tradeoffs explicitly rather than leaving them implicit in a conversation a junior engineer happened to overhear. It's also durable. A new hire eight months from now can read the same doc a current engineer wrote, which in-person osmosis never offered since it required being in the room at the right moment.

Structured onboarding buddies work reasonably well for the access problem specifically. Giving someone an explicit, sanctioned person to ask questions removes the social cost of interrupting a stranger, which is most of what made in-person interruption easy in the first place. This is a genuine fix, not a compromise, provided the buddy relationship has real time budgeted against it rather than being an unpaid addition to someone's existing workload.

Office hours work for narrow, specific questions that someone already knows they have. They work less well for the questions a junior engineer doesn't yet know how to ask, which is often the more important category early in someone's development.

Where they fall short

Recorded pairing sessions are the clearest example of a substitute that sounds good and underperforms in practice. A recording captures the words but not the improvisational, interruptible quality of live pairing. Nobody can pause a senior engineer mid-thought and ask why they just tried that particular approach. Recordings also require someone to go looking for them, and junior engineers rarely know in advance which forty-minute video contains the lesson they need. Osmosis worked partly because it was ambient. You didn't have to search for it; it happened near you. A library of recordings reverses that: the knowledge exists, but the discovery burden now sits entirely with the person who needs it most and knows the least about where to look.

Code review has absorbed a lot of the mentorship burden that used to happen in person, and research on what code review actually accomplishes suggests it was never built to carry that load alone. Academic work on modern code review practices has found that knowledge transfer is a real function of review, but a secondary one, and it competes directly with review's primary job of catching defects before merge. When teams lean on review as their main mentorship channel, the two purposes interfere with each other: a reviewer focused on finding bugs isn't necessarily explaining reasoning, and a junior engineer reading terse review comments doesn't get the why that used to come free in a live walkthrough. This dynamic gets worse, not better, once review has to clear time zones instead of a desk, a problem one piece on this site has looked at directly in why code review becomes a bottleneck once engineers span time zones.

The deepest loss is calibration, and it's the one nobody has built a good substitute for. Research on remote and hybrid work has found that distributed collaboration tends to make communication networks more static, with people talking mostly to the colleagues they already talk to and less to people outside their immediate team, which is exactly the kind of incidental cross-pollination that used to expose junior engineers to how different senior people think. A related piece on this site looked at what AI code review quietly removes from engineering teams, and the pattern is similar here: when the ambient signal disappears, nobody notices until the junior engineers who should be ready for more responsibility aren't.

Diagnosing the gap

The useful exercise for a distributed team isn't asking whether mentorship is happening. It's asking which of the four mechanisms, reasoning exposure, context transfer, low-stakes access, and calibration, has an actual owner and budgeted time, and which one everyone is quietly assuming still happens on its own. On a team of fifteen or more, calibration is usually the one nobody owns, because it was never a task to begin with. It was just what happened when people sat near each other. Distributed teams that treat it as a task, by deliberately rotating junior engineers across projects and senior engineers rather than letting pairs calcify, tend to close the gap faster than teams that just add another recorded session to the onboarding folder.

No comments yet. Be the first to comment.