Two of you, four functions, and a conversation that has been circling for a fortnight. One of you will take product and engineering, the other commercial and operations, and both of you have opinions about which way round.
You are treating this as a question with a right answer. You have almost no evidence about who is better at what, the allocation will be effectively permanent within a year, and neither of those is the thing that will damage the relationship.
Why these three models
The decision is how to divide functional responsibility between founders, and what to do about the fact that you are deciding it now. The features that fire are a judgement made on a very short record, an assignment that accumulates its own justification over time, and a division that will be evaluated for fairness rather than for optimality.
Three lenses. Luck and skill produce a random answer about how much your evidence can support. Path dependence produces a cycle answer about why the split stops being reversible. Reciprocity produces an equilibrium answer about what actually breaks. The third is the one founders least expect and the one that ends partnerships.
1. You are deciding on almost nothing
Start with the evidence available, because both of you are speaking with more confidence than it supports.
What you have is a short history in a high-variance environment. One of you closed the last three customers, which may reflect selling ability or may reflect that those three customers came through a personal contact. One of you shipped the last release, which may reflect engineering judgement or that the release was straightforward. Separating contribution from circumstance takes far more observations than a young company has produced, and short sequences in variable settings produce apparent differences routinely with nothing underneath them.1
Which means confident statements about who is better at what are mostly retrospective narrative. That is not an argument for indecision. It is an argument for holding the decision loosely, because the case for it is much weaker than the conversation implies.
What you can use instead is prior evidence rather than company evidence. What each of you did before, for years, under conditions somebody else observed. That is a longer record and it is the one thing in this decision with any sample size behind it.
Six months of company history is not a track record, it is an anecdote you were both present for.
The confidence is worth understanding rather than resisting, because it will be present in both of you. Each side is reasoning from a vivid recent episode rather than from any base of comparable cases, which is the method that produces specific-feeling conclusions and cannot see the pattern.4 You are both certain, and certainty here is a property of the story rather than of the evidence.
That prior evidence needs a form, or it turns into another story. Each of you writes, separately, the three hardest things you have done in the last five years, who else was there to see them, and what went wrong. Then you read each other’s. It takes fifteen minutes and it reaches the thing the argument never gets to: one of you spent four years managing people and it had not come up, because it happened at an employer neither of you thinks of as impressive.
Where neither of you has a record, because you are both doing this for the first time, stop hunting for one. Allocate on who wants which function more, say plainly that this is the basis, and shorten the review to six months. A preference is a weak reason and an honest one, and an honest weak reason does far less damage in year two than a manufactured strong one.
And the honest version of the residual: for most pairs, several allocations would work. The decision matters less than either of you thinks, which is a useful thing to say out loud because it lowers the stakes of a conversation that has been circling.
2. It hardens faster than you expect
The second lens explains why a provisional split stops being provisional without anybody deciding that it should.
Whoever takes a function starts accumulating things that attach to it. The customer relationships form with one person. The engineering context lives in one head. The supplier knows one name. Each of those raises the cost of swapping later, and the cost rises continuously while nobody is watching it.2
Within a year, a swap is not a conversation about preference. It is a transfer of relationships, context and credibility that costs the company a quarter, which means the provisional split you agreed in month two has become the structure of the company by month fourteen without ever being confirmed.
Two responses follow and both are cheap now and impossible later.
Put a review date on it. Twelve months, in the calendar, with an explicit statement that the allocation is provisional until then. That does not prevent the accumulation and it does create one moment where changing is normal rather than disruptive.
And deliberately share a few things across the boundary. Both of you in the largest customer’s key meetings. Both of you able to run the fundraise. Not because duplication is efficient, but because a small number of shared surfaces keeps the swap possible and, more importantly, keeps each of you legible to the other’s world.
Keep the shared surfaces to a number you can name, because the version that fails is the one where everything is shared. Two founders in every meeting is not depth, it is two people doing one job while four functions wait. Pick three and write them down: the largest customer, the fundraise, and the monthly numbers. Everything else has a single owner who reports on it.
Expect the sharing to be resisted by whoever is furthest ahead in their function, and expect the resistance to sound like efficiency, because it is. Do not argue the efficiency point, since they are right about it. Name what the sharing is actually buying: in eighteen months, when one of you is unavailable for a month, the company either continues or it stops, and that is being decided now by whether anybody else has ever been in the room.
The failure this avoids is specific and common. Two founders who divide completely become two people who cannot evaluate each other’s work, which converts every disagreement into a question of trust rather than of substance, because neither has any basis for a view.
3. The process is what gets remembered
The third lens is the one that decides whether this is a decision or a wound.
People respond to whether they were treated fairly rather than to the absolute outcome, and they will act on that judgement at a cost to themselves.3 Which means a co-founder who ends up with the function they wanted less, through a process they regard as fair, is in a completely different position from one who got the same function through a process they experienced as a decision already made.
The specific things that make it read as fair are cheap and mostly procedural. Both of you state your preference before either argues for an allocation. The criteria are named before the assignment rather than constructed to justify it. The person who gets the more visible role is not automatically the one who becomes the public face, because those are separable and conflating them is where most of the resentment comes from. And the review date exists, which converts a permanent verdict into a first arrangement.
If you are reading this after the split was already made badly, the repair is available and it has one requirement: it must be opened by whoever got the better end. A founder raising their own allocation is heard as a claim. The same conversation started by the other person is heard as a repair. So if you took the function you wanted, open it yourself, name the process fault plainly, and put the review date in the calendar during the same conversation.
The visibility question is the one to handle explicitly, because it is the real content underneath most of these arguments. In practice one of these functions produces conference invitations, investor relationships and press, and the other produces a working company. If that asymmetry is not named, the founder in the second role will experience it accumulating for two years and will describe the problem, when it finally surfaces, as something else entirely.
Say it now: one of these roles is more visible, that is not a statement about contribution, and here is how we will handle the external representation. That paragraph costs nothing at the start and cannot be introduced credibly after the imbalance exists.
Then make it a rule rather than an understanding, because understandings are not checkable. Alternate who speaks at events. Put both names on the investor update. Have the less visible founder present the numbers at the board. Three specific arrangements, none of which costs anything, and all of which somebody can verify a year from now.
What the three say together
- Weight prior experience over the last six months. The company’s record is too short to separate ability from circumstance.
- State preferences before arguing allocations, and name the criteria before the assignment.
- Put a twelve-month review in the calendar and say explicitly that the split is provisional.
- Name the visibility asymmetry now, and decide separately who represents the company externally.
Where they disagree
The evidence argument and the path argument recommend opposite postures.
The luck-skill view says hold it loosely: you know very little, several allocations would work, and the appropriate response to weak evidence is a weak commitment that you revisit when you know more.
The path-dependence view says commit hard and early. Accumulation is what makes a founder effective in a function, and a pair who treat the split as tentative will both hedge, neither will build the depth the role requires, and the company gets two half-owners of everything.
The reconciliation is to be firm about the allocation and explicit about the review. Commit fully inside the period, so that both of you accumulate properly, and put one date in the calendar at which changing is a normal option rather than a crisis. What fails is the version most pairs land on, which is a split that is never described as provisional and never confirmed either, so it hardens by default and neither founder ever agreed to it.
What none of them contain
None of the three handles the case where one of you is better at everything. That happens, it is uncomfortable, and no allocation framework resolves it. The honest conversation there is about what the second founder’s role actually is, and it is a different and harder discussion than dividing functions.
None of them prices what investors will read into the split. The person who takes commercial usually becomes the chief executive in everyone else’s mental model, which affects the next fundraise and the cap table conversation, and it happens whether or not you titled anyone.
None of them handles the pair who want the same function. The models here assume a division exists that both can accept, and sometimes two founders both want product because that is why each of them started a company. That is not an allocation problem, it is a question about whether the second founder’s role was ever defined, and dividing the functions on top of it postpones the conversation rather than settling it.
And one property the ensemble will not produce: the allocation shapes who each of you becomes. Two years running operations makes a different person from two years running product, and the founder you are choosing to be is a larger decision than the function you are choosing to run. Neither of you can evaluate that now, and it is worth saying at the review rather than at the start.
The one action that survives the ignorance: before the next conversation, both write your preference and your reason privately, then exchange them unedited. Most of these arguments are two people advocating for allocations without ever having stated plainly what they want, and the exchange frequently resolves it in ten minutes.
Who has to move
Both of you, and the process matters more than the outcome. The instinct is to argue the merits of each allocation, which is arguing about evidence that does not exist and hardens positions on both sides. The cheapest first test is the private written preference exchanged unedited, followed by one sentence agreeing the review date. Those two steps convert a fortnight of circling into a decision that neither of you will resent, which is the actual objective.
Sources and notes
- Michael J. Mauboussin, The Success Equation: Untangling Skill and Luck in Business, Sports, and Investing, Harvard Business Review Press. The luck-skill continuum, and the consequence that short sequences in high-variance settings routinely produce apparent differences with no underlying cause, underpin section 1. The claim used here is directional; no specific number of observations is asserted, because the required sample depends on the variance of the activity and almost nobody measures their own.
- Stefan Thurner, Rudolf Hanel and Peter Klimek, Introduction to the Theory of Complex Systems, Oxford University Press. Path dependence and increasing returns, in which each step raises the cost of moving away from a configuration and the order of events determines which state is reached, are developed in the chapters on evolutionary processes. Used in section 2 for the hardening of a provisional allocation into a structure nobody confirmed.
- Jeffrey Carpenter and Andrea Robbett, Game Theory and Behavior, MIT Press. Inequity aversion and reciprocity, including the experimental finding that people reject allocations they judge unfair even at a cost to themselves, are developed in the chapters on social preferences. Used in section 3 for the claim that the perceived fairness of the process governs the relationship more durably than the allocation does.
- Sanjit S. Dhami, The Foundations of Behavioral Economic Analysis, Oxford University Press. The inside view, in which a judgement is constructed from case-specific detail rather than from a reference class, explains the confidence both founders bring to a decision the evidence cannot support: each is reasoning from a vivid recent episode rather than from any base of comparable observations. Cited for the framing in section 1.
A note on a number this article does not give. There is no period after which a role split should be reviewed that transfers between companies. Twelve months is an illustration chosen because it is long enough to accumulate real depth and short enough that a swap is still affordable. What transfers is that the date should exist and be in the calendar before the split takes effect.
Joshua Agonya Pi’Rwot, Founder.