Your strongest engineer has asked to lead the team. They have earned it by every measure you have, they will leave within a year if you say no, and there is nobody else you would trust with it.
You are about to make two decisions and discuss only one. You are removing your best individual contributor and appointing a manager with no record, and the promotion letter treats those as a single event.
Why these three models
The decision is whether to move an individual contributor into management, and how to structure the move if you do. The features that fire are a role requiring responses the previous one never called for, a point at which the nature of the contribution changes discontinuously, and a commitment that is conventionally written as permanent in both directions.
Three lenses. Variety produces an equilibrium answer about what the new role actually demands. Thresholds produce a complex answer about the point being crossed and why nobody notices it. Options produce a random answer about how to keep the decision open. The third is where almost all of the available improvement sits, and it costs nothing.
1. The new job needs responses the old one never called for
The promotion is usually framed as more seniority, as though management were the top of the same ladder. It is not. It is a different regulator with a different set of responses, and the useful question is which responses the person has and which they have never had occasion to develop.
List them concretely rather than abstractly. Deciding without doing, when you can see the faster path yourself. Giving feedback that changes behaviour rather than feedback that is technically correct. Sequencing other people’s work when their estimates are wrong. Absorbing a complaint without passing it downward. Saying no to a request from you in defence of their team.
An engineer who has been excellent for four years may have had cause to develop none of those, because the role did not require them. That is not a deficiency, it is what the previous job selected for. But a regulator with fewer distinguishable responses than the situations arriving at it cannot control the system, however hard the person works, and adding effort does not add response categories.1
Which gives the first practical move. Before the conversation, write the five situations their team will generate in the next quarter, and for each, ask whether you have ever seen this person handle something like it. Not whether you believe they could. Whether you have seen it. That distinction converts a judgement about potential into an inventory of evidence, and the gaps it exposes are the training plan.
Two cautions on the evidence. A situation handled once with you in the room is weak evidence, because your presence was part of the response. Ask who would have made the call had you been travelling. Experience from a previous job counts, but only if you can name the outcome rather than the event. A reference call that asks whether someone managed well returns nothing. One that asks what they did when a delivery slipped and the client was already angry returns a response set.
The cheaper version, for a founder who cannot restructure anything this quarter, is to hand them one of the five situations while they are still an individual contributor, with no title attached. Lead on a single project for six weeks. That produces most of the same evidence and costs you nothing you cannot take back. In a fifteen-person company it is usually the only version available.
The two repairs are the same as anywhere else. Reduce the variety arriving at them, by narrowing the team’s scope or keeping certain decisions with you for two quarters. Or expand their response set deliberately, by putting them in the specific situations they have not met while you are still there to catch the outcome. Doing neither and hoping is the common version.
2. The threshold you are crossing without naming it
The second lens is about what changes at the moment of the move, and why it feels gradual and is not.
An individual contributor’s value is what they produce. A manager’s value is what their team produces, minus what the manager would have produced alone. That is a discontinuity rather than a slope: at some point the person stops being measured on output and starts being measured on other people’s output, and there is no intermediate state where both apply.2
Systems tend to absorb pressure through slack until the slack is gone, and then move all at once. Here the slack is the new manager continuing to write code, which is what almost every newly promoted engineer does, because it is the thing they know they are good at and the thing that produces visible output this week. It works for a quarter. Then the team is large enough that the management work cannot be done in the gaps, and the transition happens abruptly and badly, usually during the first difficult month.
The failure is not that they cannot manage. It is that nobody said which job they were now doing.
The instruction is to name the threshold explicitly and in writing. What proportion of their time is management. What they are now accountable for. What they are no longer accountable for, which matters more and is almost never stated. And critically, what happens to the technical work they were carrying, because a promotion that does not reassign their previous responsibilities has not moved anyone anywhere, it has added a job.
Put an arithmetic on it before the first month rather than after. A manager of four who keeps a delivery commitment of their own is carrying about a day and a half of one-to-ones, reviews and interruptions on top of a full technical load, and that fails in week three, not month six. The first symptom is not a missed deadline. It is that their own work stops moving on the two days their meetings cluster, and they make it up at night for a fortnight before anyone says anything. Watch the calendar, because the delivery board is the last place this shows.
The opposite failure reads as clean adoption and costs more. Some newly promoted engineers stop touching the code on day one, and within two quarters they cannot read a change well enough to judge it, which leaves them managing by reported status. State a floor as well as a ceiling: they still review work, they no longer own a delivery date.
3. Make it reversible, in both directions
The third lens is the cheapest intervention in this article and the one most companies never make.
The conventional promotion is irreversible by construction. It is announced, the title changes, and the only available exit is a demotion that everyone will read as a failure, which means the company will avoid it for a year past the point where it should have happened. Where a commitment is expensive to reverse and the uncertainty is high, the value of preserving flexibility is at its maximum, and both conditions hold precisely here.3
So manufacture the option. Two reports rather than eight, for one quarter, with the scope stated as a trial in the same sentence as the offer. A named date at which both of you decide again. A senior individual-contributor track that exists and is compensated comparably, so that returning is a lateral move rather than a demotion.
A company whose only ladder is management converts its best builders into average managers, one reasonable promotion at a time.
That last item is the structural one and it has to exist before the promotion, not after. A company where the only path upward is management will keep converting excellent engineers into ordinary managers, and the ones who discover they dislike it will leave rather than ask to go back, because there is nowhere to go back to.
A reversal also needs a landing place, not only a date. A trial that ends with an agreement that it was not for them, and no defined role to return to, is a resignation with two months of notice built in. Write the return role now, while the mood is generous: the same pay band, the technical scope they had, and a specific piece of work waiting for them. If you cannot name that work today, you do not have a trial. You have a promotion with a review meeting attached.
One local constraint is worth saying plainly. Where the senior engineering market is thin and salaries are quoted with the title, a return to individual contribution reads outside the company as a demotion whatever you call it internally, and it follows the person into their next interview. That is the actual reason people refuse to go back. You cannot fix the market, but you can fix the record. Agree in writing what each of you will say about the trial if it ends, then say it the same way to the team and to anyone who calls for a reference. Someone who knows the story in advance will take the trial.
Say the reversibility out loud at the start, to them and to the team. A trial framed as a trial by both parties costs nobody anything if it ends. The same arrangement, unstated and then reversed, is a public failure that a good person may not stay through.
What the three say together
- Write the five situations their team will produce next quarter. For each, ask whether you have seen this person handle something like it.
- Name the threshold in writing: what they are accountable for, what they are no longer accountable for, and who takes their technical work.
- Structure the move as a trial with a scope and a date, said out loud to both the person and the team.
- Build the senior individual-contributor track before you need it, so that returning is lateral rather than a demotion.
Where they disagree
The variety argument and the option argument recommend opposite scopes.
Variety says start narrow: two reports, restricted decisions, a controlled set of situations, so the response set can be built without the system failing while they build it. The option argument says a trial that is too small does not test anything, because the situations that reveal whether someone can manage are the difficult ones, and a two-person team in a quiet quarter will produce none of them. You will reach the review date with no evidence and renew by default.
The resolution is to control the volume and not the difficulty. Keep the team small, and deliberately route one hard situation to them rather than shielding them from it: the difficult performance conversation, the deadline that will be missed, the disagreement with another team. Then be present for the outcome. That produces evidence at low volume, which is exactly what neither argument gets on its own, and it requires the founder to resist the instinct to protect a new manager from the only thing that would tell you anything.
What none of them contain
None of the three prices what the promotion does to the people not promoted. A team of six with one obvious candidate has five people forming a view about how advancement works here, and the reasoning you give publicly matters more to them than to the person promoted.
None of them handles the case where the request is really about money or status rather than about the work. Management is often the only route to both, and a founder who solves that with a title has answered a compensation question with an organisational change. Ask what they want that they do not have now, before you assume the answer is a team.
None of them handles the founder who is themselves the missing manager. Much of what a new manager struggles with is work the founder was absorbing invisibly, and the promotion transfers a job nobody had ever written down. If you cannot describe what you were doing for that team, the person taking it over is being handed an undefined role and will be assessed against your unstated standard.
And one property the ensemble will not produce: a new manager changes what the team tells you. Information that used to reach you directly now passes through someone who is themselves being assessed, and it arrives filtered in ways neither of you intend.4 That is a permanent change to your own visibility and it starts the day the reporting line moves, not the day they become good at the job.
The one action that survives the ignorance: before the conversation, write the five situations and mark which you have actually seen them handle. If most are unmarked, the answer is not no. It is a trial with a date, a smaller scope, and one hard situation routed to them on purpose.
Who has to move
The founder, and the work is in the structure rather than in the decision. The instinct is to treat this as a yes or a no, which is the framing that makes it irreversible, because a yes announced is a yes that cannot be revisited without cost to everyone. The cheapest first test is the five-situation list, which takes twenty minutes and usually shows that the question is not whether this person can manage but which two of the five you have any evidence about.
Sources and notes
- Stafford Beer, Cybernetics and Management, English Universities Press, 1959. The law of requisite variety, attributed to Ashby, is adopted in the chapter on the Black Box as an axiom in the process of controlling complex systems, with the condition stated as sufficient permutative variety to provide a one-to-one transformation from the controlled system into the control box. The two repairs of attenuating incoming variety and amplifying the regulator’s are named as such in Brain of the Firm, Allen Lane, 1972. Used in section 1, applied to the specific response categories a management role requires rather than to a count of situations.
- Stefan Thurner, Rudolf Hanel and Peter Klimek, Introduction to the Theory of Complex Systems, Oxford University Press. Threshold behaviour, in which a system absorbs change with little visible response until a critical point and then shifts abruptly, is treated in the chapters on critical phenomena. Used in section 2 by analogy to the transition from individual to managerial contribution, where the slack absorbing the change is the new manager continuing to do the old job.
- Sanjit S. Dhami, The Foundations of Behavioral Economic Analysis, Oxford University Press, and the real-options literature it references. The value of preserving flexibility rises with the irreversibility of the commitment and with the uncertainty to be resolved. Used in section 3. Both conditions are unusually strong for a management promotion, which is conventionally structured as permanent in both directions and made on evidence about a different job.
- Jeffrey Carpenter and Andrea Robbett, Game Theory and Behavior, MIT Press. The treatment of imperfect monitoring, and of what happens to information that passes through an agent who is themselves being evaluated, supports the observation in the blind-spot section that a new reporting layer changes what reaches the founder. The filtering is structural rather than dishonest, which is why it persists after the manager becomes competent.
A note on a number this article does not give. There is no team size at which a manager becomes necessary, and no proportion of time that defines the role. Both depend on how much variety the team’s work generates, which you can list and no benchmark can supply. The five-situation exercise is the measurement.
Joshua Agonya Pi’Rwot, Founder.