Every few months a group of eminent people signs something. A letter warning about a technology, a statement of consensus on a risk, a call for a pause or a rule or a reckoning. The signatures are impressive, the coverage is wide, and a founder whose company touches the subject feels an immediate pressure to have a position, ideally the same one. Then, quietly, someone equally eminent writes a few thousand words explaining why they did not sign, and that essay gets a fraction of the attention and contains most of the useful information. The instinct to follow the headline is understandable. The discipline to read the argument is where the judgment lives.
The reason expert letters are a poor guide for operators is not that the experts are wrong. It is that the letters are written for other experts. A consensus statement is a move inside a field: it marks a boundary, claims a position, and signals to peers and funders and regulators where a group of serious people stands. It compresses a complicated set of disagreements into a form that can be signed, which means it removes precisely the nuance an operator needs to make a decision. Reading it as an instruction is like reading a treaty as a manual. The document was never trying to tell you what to do.
What the dissent tells you
This is why the refusals are worth more than the signatures. When a respected mathematician explains in public why he declined to sign a letter from his peers about the field’s future, as Timothy Gowers did recently, the explanation is forced to engage with the actual claims: which ones he agrees with, which he thinks are overstated, where the evidence is thin and where the letter’s framing smuggles in a conclusion. That is the material a founder can use, because it shows where the live uncertainty is. The letter says the field is worried. The dissent says what, specifically, the field is worried about, and how confident anyone should be.
A founder reading both should be looking for the gap between them, because the gap is the map of what is actually unknown. If the letter asserts a risk and the dissent accepts the risk but disputes the timeline, the timeline is the open question and the risk is not. If the dissent rejects the framing altogether, the framing is contested and the founder should be careful about building on it. Reading only the letter gives a founder a position. Reading the argument gives them a model, and a model is what decisions are made from.
There is a habit worth building here, which is to read the signatories as carefully as the text. Who signed tells a founder which communities are anxious. Who conspicuously did not, and what they said instead, tells a founder where the informed disagreement lies. A letter signed by everyone in a field is a fact about the field. A letter signed by two thirds of it, with the other third explaining themselves in public, is a live argument, and live arguments are where an operator’s own evidence can actually be decisive.
Consensus and your own test
None of this means consensus should be ignored. It means consensus should be located: what kind of claim is it, how far does the expertise behind it actually extend, and does it bear on the specific decision in front of you. Experts in a field are reliably right about the field. They are far less reliably right about how the field’s findings play out in a particular business, in a particular market, with particular customers, because that is not what they study. A founder whose product depends on a contested claim should treat the consensus as a strong prior and their own market as the experiment, and should run the experiment rather than citing the prior.
The letter says the field is worried. The dissent says what, specifically, the field is worried about, and how confident anyone should be.
Running the experiment usually means something small and fast: a version of the product that assumes the consensus is right, a version that assumes the dissent is right, and real customers choosing between them. The results will not settle the expert argument, which was never the point. They will tell the founder which reading of the argument describes their own customers, and that is the only reading that changes what the company should do.
The cost of deferring
There is a specific danger in citing someone else’s position as the reason for a decision, and it is that the decision then cannot be reversed without admitting the citation was wrong. A founder who says we are not building this because the experts say it is unsafe has bound the company’s roadmap to a document they did not write and cannot amend, and if the experts revise their view, or the dissent turns out to be right, the founder is left defending a position they never actually held. Decisions made on borrowed judgment are expensive to unwind because the borrowing has to be unwound first.
The same trap catches founders who cite expert consensus to justify building something. We are building this because the field says it is the future is a sentence that reads well in a deck and badly in a post-mortem, because the field was describing its own expectations, not the founder’s customers, and the founder has substituted one for the other.
Owning the decision is different. A founder who says we are not building this because we ran a test and our customers did not want it has a decision they can reverse the moment the test says otherwise, and no reputational cost in doing so. The judgment was theirs, the evidence was theirs, and the change of mind is simply a response to new evidence rather than a confession of having followed the wrong crowd. This is the practical case for reading the argument: it leaves the founder with a position they can defend and revise, instead of one they can only repeat.
Holding a contrarian read honestly
The risk on the other side is real and should be named. A founder who learns to discount expert letters can slide into discounting all expert evidence, and become the operator who ignores every warning because warnings are just signals about anxiety. That is not judgment. It is a different way of not thinking. The distinction is whether the founder is engaging with the argument or dismissing it, and the test is simple: can they state the strongest version of the position they are declining to adopt, and say precisely where and why their own evidence diverges from it.
A founder who can do that has earned the right to a contrarian read, because the read is built from the argument rather than against it. A founder who cannot is just contrarian, which is a personality rather than a method. Experts disagreeing in public is a gift to anyone willing to read the disagreement. The headline tells you what a field is feeling. The argument tells you what it knows, what it does not, and where your own market is the only source of evidence that counts.