It is the third time you have had this argument.
The services contract would cover nine months of runway. It also puts both engineers on someone else's roadmap until spring. You argued it in the car, then over Slack near midnight, then again on Sunday — and every round started at the beginning, because neither of you wrote down what the last two settled.
Most deadlocks are assumption conflicts in costume
By round three it stops feeling like a decision and starts feeling like a values conflict: one of you wants to build a product, the other wants to keep the lights on. That framing is usually wrong, and expensive — a values conflict has no procedure, and a factual disagreement has several.
Pull the positions apart and the disagreement rarely lives where it appears to. One of you assumes the diversion is reversible: engineers come back in spring, nothing structural changes. The other assumes it is not, because a paying client will want a second phase, and saying no to revenue is harder than to a prospect.
Neither of you has said that out loud. You are arguing a conclusion while holding different premises — unwinnable by construction, because every point you make follows validly from something the other never accepted, so it reads as stubbornness rather than reasoning. It compounds too: effort spent defending a position increases commitment to it.
Write down what you already agree on
Before touching the open part, list the settled part. It is almost always longer than it feels: the runway number, that the client is real, that neither of you wants to be a consultancy in two years, that the timeline slips either way.
The disagreement then shrinks to its actual size — usually one or two open items rather than the whole strategy — and the next round does not start from zero, because only what is written down survives the conversation.
State the position and the assumption under it
"I think we take the contract" is a conclusion. "I think we take the contract, because I am assuming we can hand it back in March without the client escalating" is a conclusion with a handle on it.
The second can be attacked where it is weak. The first can only be agreed with or resisted, which is why round two sounds like round one. Dropping to the premise also puts the argument where neither of you has invested any ego, and often ends things outright: once a premise is visible, one of you usually knows something about it the other does not.
Marking which claims are confirmed and which are your read matters more between cofounders than anywhere else, because shared context is why premises go unstated. You skip them assuming they are obvious. They are obvious to you.
Both of you name what would change your mind
Each person writes the evidence that would flip them. "I would take it with a written scope cap." "I would walk if their last two vendors both got extended."
That converts an argument into research, and research ends. The comparison matters more: if one of you can name a disconfirmer and the other cannot, that asymmetry is worth discussing on its own; if neither can, this is not a factual disagreement at all — skip to the last section.
Agree who decides before you need to
Pick the decider in advance. Chosen during a live deadlock it looks like a power grab; chosen beforehand it is an operating rule. Split by domain, or by who lives with the consequence.
Then say what a decision means: the company moves. It does not mean the other person was wrong, or require them to perform agreement. Keep the open question open and dated, so whoever lost the call need not pretend to be convinced. Most lasting resentment in cofounder decisions comes from being asked to agree when all that was needed was to proceed.
What labeled assumptions do that a shared doc doesn't
Handing over the session is still the wrong move — your cofounder does not need to watch you talk yourself into and out of a position four times. A briefing is that session composed for a reader.
For a deadlock the field that earns its place is assumptions, each labeled stated by sender or inferred by sender's AI — a distinction conversation cannot carry. Tagged inferred by sender's AI, "the client will want a second phase" reads as a model's reading of your session rather than a position you have committed to, so pushing back costs your cofounder nothing socially. Tagged stated by sender, they know you own it. Every claim in an argument arrives at the same volume. Here they arrive sorted, and the reply lands on the same thread.
Where structure will not help
All of this resolves assumption conflicts. It does nothing for the other kind.
If you genuinely disagree about what the company is for, no document dissolves that. You will write the agreed list and find it is all logistics; you will name your disconfirmers and find neither of you has one, because the disagreement was never about evidence. That conversation is worth having on its own, and may need a tiebreaker outside the two of you.
Structure still earns its keep: it tells you which argument you are in, in twenty minutes rather than three months. Do not expect your assistant to break the tie either — it will argue whichever side you ask it to, which is exactly why it cannot. The opinions it hands you are not positions until you decide they are.
Five worked examples, including a cofounder decision, are in AI briefing examples. If your last three arguments were the same argument, start with the agreed list. Connect your assistant when you want it composed from the session.