Rebuilding Trust With Your Cofounder After a Fallout
Something specific happened, and now you are both being professional at each other. The company runs, the meetings happen, and something load-bearing has gone out of the relationship. What makes this stage confusing is that the obvious repair — talking about it, apologising, agreeing to move on — usually produces a few good weeks and then a return to the same distance. That is not because either of you is insincere. It is because trust is not restored by conversation. This page is about what actually rebuilds it, and how to tell repair from a slow ending.
What’s inside
Trust is a prediction, not a feeling
Which of the three trusts broke
The repair sequence
Why apologies are not the mechanism
The fortnight that decides it
What unrepairable actually looks like
Where Blomma fits
Trust is a prediction, not a feeling
The useful definition: trust is your working forecast of what someone will do when you are not watching. It is built from accumulated evidence, and it is revised by new evidence.
That definition explains why the standard repair fails. A conversation, however honest, adds no evidence about future behaviour — it adds a stated intention, which is a different and much weaker input. Your forecast does not update because someone told you it should. It updates when you observe the thing they said they would do, several times, including once under pressure.
It also explains why the repair feels slower than it seems it should. You are not waiting to feel better. You are waiting to accumulate enough observations to revise a prediction that was formed over years and then contradicted. That takes a specific number of instances, not a specific amount of goodwill.
So treat this mechanically. What prediction did the rupture change? What evidence would revise it? Then set up conditions where that evidence can accumulate. Founders who approach repair as an emotional project stall; founders who approach it as an evidence problem tend to get somewhere.
Which of the three trusts broke
Three distinct kinds, frequently conflated, with different repairs. Getting this wrong is why so much repair effort goes nowhere.
Reliability.Will they do what they said? Broken by missed commitments, things dropped, follow-through that did not happen. The most repairable of the three, because the evidence is unambiguous and accumulates quickly — a handful of kept commitments genuinely moves the forecast.
Judgment.Will they make good calls in their domain? Broken by a significant decision that turned out badly, or by discovering they were operating on reasoning you consider unsound. Repairable, more slowly, and the repair is not reassurance — it is watching them make calls in a bounded area and seeing the reasoning. Note that this one is often confused with a capability question, which is a different conversation entirely.
Intent.Are they on my side? Broken by something that read as self-interested at the company’s expense, or at yours — going around you, a private conversation with an investor, a decision that advantaged them. The hardest to repair and the only one that is sometimes genuinely terminal, because it changes the interpretation of everything else. Once intent is in question, every ordinary act gets read for motive, which generates fresh evidence continuously.
Be precise about which one you are dealing with. Founders often say “I don’t trust them any more” when the accurate statement is “I don’t trust them to follow through on operational commitments,” which is a considerably smaller and more tractable problem.
The repair sequence
Five steps, in order. The order matters more than any individual step.
One. Establish what happened, factually, before anyone interprets it.The sequence of events, who knew what, what was said. Less contested than either of you expects, and agreeing on it is the only footing the rest has. Skipping to meaning is what turns repair into an argument about motive.
Two. Name the prediction that changed.Specifically. “I no longer assume you’ll tell me before making a commitment to a customer” is workable. “I don’t trust you” gives neither of you anything to act on.
Three. Identify the evidence that would revise it.Concrete and observable. If the broken trust is reliability, it is a set of kept commitments. If it is judgment, it is decisions made in a bounded area with visible reasoning. If it is intent, it is behaviour under a genuine conflict of interest where they choose against their own advantage — which cannot be manufactured and has to be waited for.
Four. Create conditions where the evidence can accumulate.This usually means a structural change: a weekly founder session, a written record of commitments, explicit decision rights, or more information flowing between you. Repair without a structural change relies on both of you behaving differently through will alone, which does not hold past the first hard quarter.
Five. Set a review point and use it.Six weeks. Not to relitigate, but to ask whether the forecast has moved. Naming this in advance prevents the alternative, which is one of you silently deciding it has not worked.
Why apologies are not the mechanism
Apologies matter and they do a different job than people think. Worth being clear about it, because founders often treat the apology as the repair and are then confused when nothing changes.
An apology’s function is to establish that both of you have the same account of what happened and that the person who caused the rupture recognises it. That is genuinely necessary — repair cannot start while the two of you disagree about whether there was an injury. What an apology does not do is provide evidence about future behaviour, which is the actual currency.
A few practical notes. An apology that includes an explanation usually reads as a defence, and the explanation cancels the apology. “I’m sorry, I was under a lot of pressure” lands as “I’m not sorry.” Say the thing you did, acknowledge the effect, and stop.
And an apology for the wrong injury does damage. If your cofounder is upset that you made a decision without them and you apologise for how you communicated it, you have demonstrated that you still do not understand the objection. Find out what the injury actually was before you address it.
The fortnight that decides it
Repair is decided in a specific window, and it is not the conversation.
About two weeks in, something will happen that looks exactly like the original rupture. A commitment will be nearly missed, a decision will be made slightly too independently, a message will land ambiguously. It may well be innocent — the base rate of ordinary friction has not changed.
What happens next determines everything. If the injured founder treats it as confirmation and withdraws, the repair ends there, silently, and both of you will later date the failure to the wrong moment. If they raise it immediately, small and specific — “this is the thing we talked about, can you flag these earlier” — and it gets handled, that single exchange does more for the forecast than the original conversation did. It is the first piece of real evidence.
So agree in advance that this will happen and how you will handle it. “When I see something that looks like the old pattern, I’m going to say so within a day rather than sit on it.” That commitment is the highest-value thing to come out of a repair conversation, because it converts the inevitable relapse from a verdict into a test that can be passed.
What unrepairable actually looks like
Sometimes it is over, and knowing the markers saves years.
Repeated repair with no structural change. If you have had this conversation three times and nothing about how you work has changed, you are not repairing — you are periodically reconciling, which feels productive and holds nothing.
The injury keeps changing. Each time you address one thing, another older thing surfaces. That usually means the specific rupture was not the problem; there is an accumulated store, and the repair needs to be a full renegotiation rather than a fix.
One of you is no longer trying to be understood. When someone stops explaining themselves — not angry, just done — that is a more reliable signal than any argument. Conflict means engagement. Withdrawal is the ending.
And intent trust that will not move despite good evidence. If your cofounder has behaved well for six months and you still read every action for motive, the forecast is not going to revise. That is information, and it is not a moral failing on either side.
In those cases the honest next step is renegotiating the arrangement or parting ways, not another attempt at the same conversation.
Where Blomma fits
The difficulty here is that you cannot assess your own forecast from inside it, and everyone you might ask is compromised. Your team can see the frost and cannot discuss it. Mutual friends will take a side. And your own read is distorted by the fact that the rupture is more vivid to you than the six ordinary months around it.
Blomma is an always-on AI career coach with no stake in the relationship. Use it to work out which of the three trusts actually broke, which is the diagnosis founders most often get wrong and the one that determines everything downstream. Use it to name the specific prediction that changed and the evidence that would revise it — concretely enough to be observable. Use it to prepare an apology that does not cancel itself, if you are the one who caused the rupture. And use it in the fortnight that decides it, at the moment something looks like the old pattern and you have to choose between raising it and withdrawing. Where it warrants a human, or where you need a third party in the room, you can bring one in..
Cofounder trust does get rebuilt, routinely, including after ruptures that felt terminal at the time. What it takes is not more goodwill. It is a precise account of what broke, a structural change that lets new evidence accumulate, and both of you handling the first relapse well.