About two years ago, someone left one of the engineering teams in my org whose contribution I didn’t have full visibility into until after they were gone. Not dramatically underappreciated — they were well-compensated, well-reviewed, treated respectfully. But I hadn’t fully seen what they were doing, and when they left, the teams they’d been quietly supporting revealed the size of the gap in the way that only becomes fully legible in retrospect.
This person was not in every meeting. They weren’t the loudest voice in planning sessions. They didn’t send the most emails or generate the most Slack messages. What they did, I understood only in retrospect, was absorb and resolve a remarkable amount of organizational friction before it could become someone else’s problem.
When two teams had a misunderstanding about an API contract, this person had usually already talked to both sides before it needed a formal escalation. When a newer engineer was confused about why a decision had been made, this person had the historical context and knew how to explain it without making the newer engineer feel bad for not knowing. When there was tension between two teams about priorities — the kind of tension that, left unaddressed, turns into a meeting with eight people and someone taking notes — this person would have a conversation, informally, that drained enough heat from the situation that the meeting never got scheduled.
None of that work was visible in a performance review template. It didn’t show up in delivery metrics. It wasn’t on any roadmap. The only way you could see it was by watching the pattern of conflicts that didn’t happen, meetings that didn’t get called, escalations that somehow never materialized. Which is, of course, almost impossible to measure, because you’d need a counterfactual.
This is one of the most persistent problems in how engineering organizations calibrate talent. We’re reasonably good at measuring output — code shipped, features delivered, incidents resolved — and much worse at measuring friction absorbed and trust maintained, because those things don’t leave artifacts. And in calibration cycles, where you’re making comparative judgments across a group of people with limited time, the things that leave artifacts tend to win.
The person who shipped a large feature wins over the person who kept three cross-team relationships healthy enough that a large feature could be shipped at all. Not because the first person is more valuable — often they’re not — but because the first person’s value is legible and the second person’s isn’t.
I’ve started to look for specific signals. The quiet leader tends to be the person other people name when I ask “who do you go to when you’re stuck on something that isn’t a technical problem?” They’re often the person who gets CC’d on emails that technically don’t require them but where someone wants their eyes on it anyway. In retrospect, the person I described earlier was on every difficult email chain in a way that I hadn’t consciously noticed until they were gone and the chains started coming directly to me or to my managers without the usual pre-processing.
What I try to do now is make the invisible work legible — not just at calibration time, but throughout the year. When I notice that someone has diffused a situation, I name it in our one-on-one. When someone’s judgment or relationship capital has saved the team from a significant coordination cost, I write it down and I look for moments to reflect it back to others. Part of this is fairness. Part of it is self-interest: if I’m not making this work visible, I’m creating conditions where the people who do it feel unrecognized and eventually leave, taking the work with them.
The other thing I try to do is be direct in retention conversations with this type of person. They’re often not the ones who tell you they’re unhappy. They tend to be, almost by temperament, the people who manage tension quietly — including their own. By the time they’ve made the decision to leave, they’ve usually been sitting with it for a while, and a conversation about compensation or scope isn’t going to change the calculus. The time to have the conversation is when there are still things to offer, not when you’re trying to reverse a decision that’s already been made.
In the case I described, the signals were there — quieter than usual, less engaged in certain conversations — and I read them as stress rather than the early stages of a departure decision. The lesson: when the signals are ambiguous, ask directly. By the time the pattern becomes unambiguous, the decision is usually already made.
The lesson isn’t complicated, but it requires ongoing vigilance: look for what doesn’t happen, ask who made it not happen, and tell that person you see it.