The first time I got this badly wrong, I kept a senior engineer for fourteen months longer than I should have because his throughput was extraordinary. He shipped the payment reconciliation engine ahead of schedule. He also made two junior engineers cry, told a product manager her roadmap was "for children," and quietly convinced half the team that code review was optional if you were good enough. I told myself the output was worth it. It wasn't. When he finally left, three other people I actually wanted to keep left within a quarter, and two of them told me on the way out that he was the reason.
So I have opinions here, and they are not the balanced, on-the-one-hand opinions you get from people who have never had to sign the departure paperwork. The toxic high performer is the single most expensive personnel problem a leader will face, precisely because the cost is hidden and the benefit is loud. This is how I think about it now.
What Actually Counts as Toxic
Let's be precise, because the word gets thrown around lazily. A toxic high performer is not someone who disagrees with you in a meeting, or who is blunt, or who has strong opinions about your database choice. I want people like that. Bluntness is not toxicity. Being difficult to manage is not the same as being corrosive to the people around you.
Toxicity is a pattern of behavior that degrades the people nearby. It is the engineer who is charming to you and contemptuous to anyone junior. It is the person who hoards knowledge so they stay indispensable, who takes credit sideways and blame downward, who makes meetings tense in a way you can feel in your shoulders. The tell is not one bad day. The tell is that a whole team gets quieter, slower, and more careful when this person is in the room, and nobody can quite explain why on the record.
The Math Nobody Does
Leaders keep these people because they do the math wrong. They compare one person's output to zero, as if the alternative to keeping the star is an empty chair. That is not the alternative. The real comparison is the star's output minus the drag they impose on everyone else.
Here is the drag I have actually measured, more or less, over the years:
- Two or three teammates operating at maybe 70 percent because they spend energy managing around one person's moods.
- Code reviews that turn into standoffs, so PRs sit for days instead of hours.
- The quiet ones who stop proposing ideas because the last three got mocked.
- Attrition. The good people have options and they use them, and replacing a mid-level engineer runs six figures once you count recruiting, ramp time, and lost context.
Do that arithmetic honestly and the "irreplaceable" 10x engineer is frequently a net 3x once you subtract what they cost the six people around them. In a regulated payments shop, where a single sloppy change to a settlement path is a real financial and compliance event, I would rather have a steady 5x who makes everyone else 20 percent better. That is not a soft-skills nicety. That is the throughput of the whole system.
Name the Behavior, Not the Person
When you finally decide to act, the first conversation matters enormously, and most managers botch it by being vague. They say things like "there have been some concerns about your collaboration style." That means nothing. The person hears it as office politics and files it away as your problem, not theirs.
Be specific to the point of discomfort. "On Tuesday in the architecture review you interrupted Priya four times and told her the design was, quote, embarrassing. That is not acceptable here, and it needs to stop." Name the incident, the impact, and the standard. You are not diagnosing their character. You are describing an event and drawing a line. High performers, whatever else is true of them, tend to respond to precision far better than to feelings-talk, because precision is a language they respect.
The Genius Exemption Is Poison
The most damaging thing a leader does is not keeping a toxic person. It is visibly exempting them from the rules everyone else follows. The moment your team sees that being brilliant buys you a pass on how you treat people, you have taught them exactly what the organization values, and it is not each other.
Every standard you enforce is only as strong as your willingness to enforce it against your most valuable person. Enforce it everywhere except there, and you have no standard. You have a suggestion.
I have watched this rot a culture from the inside. Once the exemption is granted, your steadiest people start wondering why they bother being decent. Some of them conclude that the way to get ahead is to become a little more like the difficult star. That is the real contagion, and it is far worse than one bad apple, because now you are growing your own.
Enjoying this article?
Get more like it in your inbox — practical engineering leadership, fintech, and AI. No spam, unsubscribe anytime.
Give a Real Runway, Then Mean It
I am not in the camp that says fire on sight. People can change, and I have seen it happen, though less often than optimists claim. So give a genuine chance: clear behavioral expectations, written down, with a timeline, and real coaching support if the person seems capable of self-awareness. Thirty to sixty days is usually enough to see whether someone is willing to work on it or merely willing to say they will.
The word that matters is "mean it." A performance conversation that has no consequence behind it is worse than no conversation, because now the person knows exactly how empty your threats are. If you say the behavior has to change or they cannot stay, you have to be genuinely prepared to lose them. If you are not prepared to lose them, do not have the conversation yet, because they will smell the bluff and you will have spent your credibility for nothing.
Watch for the Charm Offensive
When you finally confront a truly manipulative high performer, a common move is the sudden pivot to charm. The difficult behavior stops overnight, they are warm and cooperative in every meeting you attend, and you start to wonder if you overreacted. Be careful. There is a real version of this, where someone genuinely takes the feedback and changes. There is also a performance of it, where the behavior simply goes underground and reappears the moment your back is turned.
The way to tell them apart is to ask the people who are not in the room with you. Go to the junior engineers, the QA folks, the PM who bore the brunt of it, and ask them plainly whether things are actually better in the meetings you don't attend. Their answer, not the star's improved demeanor toward you, is the signal. If the behavior only improved in your presence, you have your answer, and it is not a good one.
When You Decide to Let Them Go
Once you have made the call, move with more speed than feels comfortable and more grace than the person probably earned. Slow, drawn-out exits are cruel to everyone and they let the departing person poison a few more relationships on the way out. Reduce their access to sensitive systems on the same timeline you would for anyone leaving a payments company, not because they are a villain, but because that is simply the control you owe the business.
Do not badmouth them to the team afterward, and resist the urge to explain yourself in detail. A calm "we've decided to part ways, and I want to make sure the reconciliation work has a clean handoff" is plenty. Here is the part that surprises new managers: the relief is almost always immediate and visible. I have had teams where the energy in standup changed within a week. People start talking again. Someone floats a half-formed idea and nobody stomps on it. That change tells you everything about what the person was costing you that no spreadsheet ever did.
The Knowledge-Hostage Problem
The thing that keeps leaders paralyzed is fear: this person is the only one who understands the settlement engine, and if they walk, we are exposed. That fear is legitimate, and it is also a diagnosis of a management failure you should fix regardless of this one person. If a single individual can hold a regulated payments platform hostage, you have a bus-factor problem that will hurt you eventually anyway, whether through resignation, illness, or a competitor's offer.
So start the de-risking before you start the exit conversation. Pair someone alongside them for a few weeks. Get the tribal knowledge into runbooks and diagrams. Yes, a manipulative person may sense what is happening and resist, but you can frame it as ordinary operational resilience, because it is. Once the knowledge is no longer hostage, the decision becomes much easier, and honestly, the person's hold over you evaporates, which is often what surfaces the worst of the behavior anyway.

Conclusion
Here is what I wish someone had told me before that fourteen-month mistake: your best people are watching how you handle your most difficult one, and they are drawing conclusions about whether this is a place worth staying. The toxic high performer is not really a talent problem. It is a test of whether your stated values survive contact with someone who is genuinely useful. Fail that test and the values were never real, just a poster. Pass it, publicly and calmly, and you will find your quieter, kinder, quietly excellent people trust you in a way that outlasts any single quarter's velocity. Protect them. They are the ones who compound.
Get new posts in your inbox
Occasional, practical notes on engineering leadership, fintech, and building with AI. No spam, unsubscribe anytime.
Comments (0)
Leave a Comment
No comments yet. Be the first to comment!

