Essay · Applied for Operators

When to Fire, When to Mentor

Autumn 2026 · Reading time · 16 minutes

The question is never whether the person in front of you is failing. Everyone in a demanding role is failing at something on any given week. The question is whether the failure is the kind that responds to being seen or the kind that hardens under attention. That distinction is the difference between mentoring and firing, and no operator gets it right by feel.

Every working owner has a private drawer of these decisions. The person who is almost good enough. The person who used to be excellent and is now only present. The person whose failures are all recoverable and the person whose successes are all accidental. Nobody teaches you how to tell them apart. Every management book pretends the answer is a process. Every senior operator I trust knows the answer is a judgment, and every serious judgment I have made in this space has run through the same four questions, in the same order, before I let myself act.

The false urgency that ruins the call

Almost every mentor-or-exit decision I have seen mishandled was mishandled under a false sense of urgency. Something happened, the owner’s calendar was already full, and the pressure to resolve the situation quickly became indistinguishable from the pressure to resolve it correctly. The two are not the same. Speed of resolution is a real value in operations, but it is a lethal value in people decisions, because a wrong call made quickly is not just wasted effort. It is trust damage across the whole team, and trust damage compounds in a way that no forensic explanation can later repair.

The rule I try to keep is that a mentor-or-exit call gets, at minimum, one weekend between the trigger and the action. Not because I need the days for information. I usually have the information already. I need the days for the situation to stop being an event and start being a pattern I can look at coolly. Events distort. Patterns clarify. The weekend is what turns one into the other. Every call I have made without the weekend has been either the right call for the wrong reasons or the wrong call altogether. Neither outcome is worth the two days I would have gained by moving faster.

Question one · is the failure local or structural

A local failure is a specific miss. The person shipped the wrong version. The person misread the room. The person made a bet that did not pay. Local failures are almost always coachable, because a coachable person can point to the moment the miss happened, describe what she was thinking at that moment, and articulate what she would think next time. A structural failure is different. A structural failure is a pattern of misses whose common ancestor is the way the person perceives the situation. Structural failures are not moments. They are lenses. And lenses are almost never coachable inside a normal working relationship, because the person cannot see the lens she is looking through.

The test I use is small. I ask the person to describe the last three misses in her own words. If she describes three specific moments, I am usually looking at a local pattern I can work with. If she describes three general narratives with the same emotional shape, I am usually looking at a lens, and the mentoring conversation I am about to have will produce sincere agreement, honest effort, and no change. Mentoring a lens is one of the most expensive things an owner can do, because the person is not resisting. She is trying. She simply cannot see the thing that would need to change. The kindest thing an owner can do at that point is not another coaching cycle. It is an honest exit.

The version of the question owners get wrong

The most common version of the local-versus-structural mistake I see is the owner who mistakes a pattern of local failures for a structural one. She has watched the person miss three times in a row, the frustration has accumulated, and by the time she runs the diagnostic she is already reading a lens where a series of local misses would be a more accurate read. Structural framing is emotionally cleaner. It gives the owner a reason to stop investing. Local framing is more work, because it requires her to keep coaching. The bias in a busy quarter is always toward the cleaner framing. That bias has cost me at least two hires I should have kept.

The correction I have adopted is a simple one. When I suspect a structural pattern, I make myself wait one more cycle and coach specifically to the pattern I think I am seeing. Not to the misses. To the pattern. If the pattern actually is a lens, the coaching will land nowhere. If it is a series of local misses that happens to look like a lens under stress, the coaching will land somewhere small and the pattern will shift within a fortnight. That extra cycle is uncomfortable and it is worth it. It has recovered adults I would otherwise have exited, and it has confirmed exits I would otherwise have delayed. Either way it makes the eventual call cleaner.

Question two · is the person the source or the site

Before firing anyone I make myself answer this one plainly. Is the failure originating with this person, or is this person simply the visible site where a larger failure is landing. A team can produce a person who looks like the problem when the problem is actually a role that was scoped poorly, a manager above her who is unclear, a tool she was given without training, or a promise made to a customer that no one in her seat could ever have kept. Owners who skip this question fire the wrong person quietly for years and never notice that the seat keeps producing the same crisis with different names. The seat is the source. The person is the site.

There is a modern version of this question that did not exist five years ago. Sometimes the failure is originating with the machine work the person is being asked to review, not with the person’s underlying judgment. An owner who has quietly put a human editor on top of a fast-moving AI process, without acknowledging that the editor’s job has changed shape, will interpret the editor’s slowness as poor performance. It is not poor performance. It is the correct pace at which a competent adult can catch the specific errors that particular pipeline is prone to. Fire the editor, and the errors go straight into the customer’s hands. The question is not always about the person. It is often about the system the person is sitting inside.

The seat that manufactures failure

Some seats are structured in a way that will produce the appearance of failure in whoever sits in them. Too many stakeholders. Too much responsibility without corresponding authority. A workload sized for two adults. A metric that punishes the trade-off the seat requires. Owners who fire the person in that seat and hire a replacement, without redesigning the seat, get to watch the same pattern play out with the new hire, usually on a slightly shorter timeline. This is one of the most expensive kinds of mistake a growing business can make, because each cycle costs a hire, a runway, and a small piece of the team’s trust that the seat can be won.

The signal that a seat is manufacturing failure is when two competent adults in a row have failed to succeed in it. That is the moment the owner has to stop evaluating individuals and start evaluating the design. Not always. Sometimes the second failure really is another individual issue. But the owner who does not at least open the design question after the second failure will almost certainly be wrong by the third, and by that point she will have exited three people who might have thrived in a seat that had been repaired between hires.

Question three · has the person’s role outrun her

This is the most common cause of a firing that everyone in the room feels bad about. A person who was excellent at a role in one season is not excellent at the same role two seasons later, because the role itself has changed shape underneath her. The company grew, the tools changed, the standard rose, the counterparties got more sophisticated. Nothing about the person got worse. The demand on the seat got higher. This is not a coaching problem and it is not a firing problem in the traditional sense. It is a re-scoping problem, and it is almost always the owner who has failed to see it in time.

In the current season, the demand on almost every professional seat is being raised by AI-augmented peers and competitors moving faster with less. A person who was a strong solo operator eighteen months ago is now, whether the owner has named it or not, in a role where the expected output is the output of one adult plus at least one competent tool. When the person has not adopted the tool, the gap looks like a performance failure. It is not. It is a training gap the owner allowed to open. The honest response is not to fire her. It is to name the gap out loud, provide the training, and give a short, specific window for the adoption to become visible. If it does not, then the question moves back to question one. But the owner does not get to skip question three.

The window that decides everything

The re-scoping conversation that follows a role outrunning a person requires a specific window, and owners who miss the window pay for it. The window opens when the gap first becomes visible and closes when the person has begun to feel it as a personal failing. If the owner names the gap while it is still a structural fact, the conversation is a partnership. If the owner waits until the person has been quietly failing for a quarter, the same conversation lands as an accusation. Same words, same intent, entirely different landing. The window is often only a few weeks wide and it does not extend for the owner’s convenience.

What closes the window fastest, in the current season, is a peer who has adopted the tools the person has not. The moment a team member sees a colleague producing three times the output with the same title, the psychological environment around her shifts, and no amount of coaching from the owner will fully repair it. That is why the training conversation has to happen at the earliest possible moment the owner suspects the gap is opening. Not the moment the numbers confirm it. The moment the shape of it becomes visible. Waiting for confirmation is the same as waiting for damage. The confirmation is the damage.

Question four · what am I telling myself I am doing versus what I am doing

This is the question I ask myself alone, and it is the one that has changed my behavior most. Before any firing conversation, I write down two sentences. The first is what I would say to my most trusted advisor about why this person needs to go. The second is what I would say if I had to be honest that the reason is at least partly about me. Am I firing this person because the seat requires it, or because keeping her would require me to have a conversation I have been avoiding for a year. Am I firing her because the customer is suffering, or because her presence in the room reminds me of a decision I made and would rather not revisit. If the second sentence is longer than the first, I am not ready to fire.

The mirror version of this applies to mentoring. Am I mentoring this person because a specific, visible, coachable pattern is present, or because letting her go would cost me a story I have been telling about my culture. Owners often keep the wrong person for the same reason they hire the wrong person. The keeping and the hiring are both flattering to the owner’s self-image, and both are paid for by everyone else in the room. Question four is the private one. It is where an owner earns the right to make the public decision cleanly.

The conversation that has to happen before either call

There is one more conversation I have learned to insist on before I do anything, mentor or exit, and it is not with the person in question. It is with the two or three adults who work most closely with her. Not to build a case. Not to gather ammunition. To calibrate. What I am checking is whether my read of the pattern matches theirs. If it matches, I move. If it does not, I have to explain the divergence to myself before I take any action. Sometimes the divergence means I am seeing something they are missing, and the movement is still the right call. Sometimes the divergence means I have been in a private frustration loop that no one else recognizes, and the movement would be wrong.

This calibration is now easier and more dangerous than it used to be. Easier, because a well-designed prompt to a well-briefed model can quickly surface counter-arguments to my own read. Dangerous, because the model will produce counter-arguments even when none actually exist in the room, and a fluent counter-argument feels like evidence when it is nothing of the kind. So I use the tool as a rehearsal partner and I use the humans as the actual check. The order matters. The tool sharpens my language before I go into the human conversations. The humans tell me whether the pattern is real. I do not let those two steps swap places.

What a clean call looks like from the outside

When an owner runs these four questions honestly, the eventual decision, mentor or exit, has a specific quality from the outside. It is quiet. It is not surprising to anyone who has been paying attention. It does not require a long justification. The person on the receiving end may not agree with the decision, but she almost always recognizes its consistency with what she has been hearing for weeks. Clean calls do not need to be explained twice. Muddy calls do. The mud is the residue of skipped questions.

There is a final thing that shows up when an owner starts running these questions across her team, and it is worth naming. She begins to make fewer of both kinds of call. Fewer firings, because she stops confusing a role outrunning a person with the person being wrong. Fewer coaching cycles, because she stops trying to mentor lenses that will not shift. Her team feels the difference before she can articulate it. The rate of movement slows and the quality of movement rises. That is the mark of an owner who has stopped optimizing for feeling decisive and started optimizing for being right. It is a slower way to lead. It is also the only way, in a season this fast, to keep the team you actually meant to build.

What the team learns from every call you make

Every mentor-or-exit decision the owner makes teaches the rest of the team something. Whether the owner intends the lesson or not. Owners who fire quickly teach the team that patience is not a value the company rewards. Owners who mentor forever teach the team that accountability is negotiable. Owners who make clean calls in either direction, at a considered pace, teach the team that both grace and standards are real, and that neither is the sole operating principle. The lesson is transmitted regardless. The only question is whether the owner has thought carefully about what she is teaching.

That is a heavier consideration than most owners give the decision. The person on the receiving end of the call is not the only audience. Everyone else in the company is watching to see what kind of owner she is now. Not what kind she was two years ago. What kind she is this quarter. Owners who forget this weight tend to make calls that feel private and land publicly, and the public landing is often the more consequential half. A call that is right for the individual but poorly weighed for the audience can damage a team’s trust in the seat far more than a slower call that took the audience seriously.

A short instruction to the owner who is deciding this week

If you are about to have one of these conversations in the next seven days, write out your answers to the four questions before you write the message. Not in your head. On paper, or in the file no one else reads. If any answer is missing, delay the conversation until it is written. The delay is not weakness. It is the smallest possible investment in making sure the version of you that walks into the room is the version that spent a private hour thinking, not the version that spent a public week reacting. Every difficult call I have gotten wrong I got wrong because I skipped that hour. Every difficult call I got right, I got right because I did not.