The Price of Disagreeing

The design review that worries me is the one that finishes early. Somebody proposes an approach, somebody else states a position with complete confidence and a reason that sounds sufficient, and the room moves on. No visible conflict, no long thread afterward, a decision in eleven minutes. I have walked out of that meeting pleased with how efficient the team was, written the outcome down as agreed, and been wrong about what I had just watched.

Here is the definition I keep coming back to, because it is the whole subject in two sentences. Consensus is what a room produces when disagreement is cheap. Silence is what the same room produces when disagreement is expensive, and in the meeting notes the two are indistinguishable.

This person is usually right, and hard to argue with. The most common version, and the least recognized, because the outcomes keep confirming the room's habit. This person's track record is real. The trouble is that a good record is being used as a general-purpose argument, so the team defers on the subjects where the record does not apply as readily as on the ones where it does. Nothing is wrong with the person. What is wrong is that the process has substituted a proxy for evidence, and the proxy happens to be accurate often enough that nobody questions it.

This person is confidently wrong, at least some of the time. Harder on the team and easier to fix, because it fails in public. The mechanism is identical: certainty is doing the work that evidence should be doing. The thing to resist is the urge to make this about calibration as a personal trait, because you will be arguing about somebody's self-image rather than about a caching strategy. It is a process problem too, and the process fix and the fix for the first case are the same fix.

This person is right, and corrosive about it. Only this one is a behavior problem. The distinguishing test is not volume or bluntness, both of which I can work with and some of the best engineers I have known have plenty of. It is whether other people are worse at their jobs in this person's presence: quieter, slower to raise a half-formed idea, routing around a review. That is a different conversation with a different owner, and dressing it up as a technical disagreement is a way of not having it.

Disagreement has a price, the price is paid by whoever disagrees, and the person who is hardest to disagree with sets it for everybody. That is not a metaphor. If contradicting someone reliably costs twenty minutes of argument, some preparation you did not plan for, and a slightly worse relationship with a colleague you need next week, then everyone on the team is running that arithmetic before they open their mouth, every time, whether or not they would describe it that way.

The oldest experimental hook for this is Asch's conformity work from 1951, where subjects asked which of three lines matched a reference line went along with a unanimous group giving an obviously wrong answer about a third of the time. I would hold it loosely: 123 male college students in the 1955 write-up, one country, one decade, and a 2015 review of textbooks found the results are routinely reported as more damning than they were, since most responses across the trials were independent.

The finding I would actually keep is the other one, which is about the fix rather than the failure. Give the subject a single other person who disagrees with the majority and the conformity largely collapses. Not a better argument, not more evidence, not authority. One other voice, which is a much cheaper intervention than anything in a management book and points straight at what a manager can do in a room without changing anybody's personality.

Set the criteria before the discussion, not during it. Write down what would make a proposal the right one - the constraints it has to satisfy, what we are optimizing, what we are willing to give up - and do it before anyone has a proposal on the table. Then the argument is about whether a design meets a stated bar, which is checkable, instead of about who is more certain, which is not. It also removes the most exhausting move available to a strong opinion, which is inventing a new criterion once its preferred option is losing on the old one.

Write the decision down with its reasoning. Not the outcome, which everyone remembers wrongly within a month, but what was decided, what was rejected, and why. A record makes both failure modes visible: being overruled by evidence looks like a reason in the document, and being overruled by volume looks like an empty line where the reason should be. That asymmetry does more work than any meeting rule, because nobody has to accuse anybody of anything. The blank space does it.

Give the opinion somewhere to go. A dissent that is written down - one line, the objection and the name attached, in the same document as the decision - is a dissent that does not have to be repeated in every subsequent meeting, and repetition is where most of the damage lives. People do not argue endlessly because they enjoy it. They argue endlessly because they are not sure they were heard, and the only evidence available to them that they were heard is whether the objection keeps needing to be said.

The adjacent research is worth citing with its limits attached. Edmondson's 1999 work in Administrative Science Quarterly gave the field a seven-item survey scale for whether people believe it is safe to take a risk on their team. The most instructive finding in that line of research is the counterintuitive one from hospital units, where a climate in which people are not punished correlates with more errors being reported. That is a claim about detection rather than incidence, and if your team gets healthier, expect more problems surfacing rather than fewer.

My job is not to have the best technical opinion. I have opinions and I use them, and the discipline is knowing that mine lands differently because of where it comes from. A manager's guess enters the room already weighted, so spending it on a caching strategy is spending it badly. The times I have had the deciding technical view and been right, I still made the room slightly worse at deciding the next thing without me, which was the harder half of a change I have written about elsewhere.

It is to notice when the quiet people stopped talking. This is the actual job and it is mostly attention rather than action. Who used to push back and does not any more. Which subjects nobody raises. Whether the objections I hear in one-on-ones are objections that were also said in the room, because that gap is the most reliable instrument I have, and it does not show up in any process metric anyone tracks.

If somebody is making their colleagues worse at their jobs, that is a performance problem in the part of the job that is about working with other people, and it should be handled the way any performance problem is handled: early, directly, specifically, and by me. Treating it as a technical disagreement to be mediated is a way of avoiding a conversation I owe them, and I have avoided it, usually by telling myself that the person is too valuable and the work too good to risk a difficult exchange.

That reasoning is wrong in a way that took me a long time to see. The cost was already being paid, just not by me and not visibly: in the review comments people stopped writing, in the design somebody stopped proposing, in the person who quietly decided this was not the team for them and gave a tidier reason on the way out. I was not avoiding a cost. I was choosing who paid it, and I chose the people who were not in the room.

Teams that optimize for comfort make worse decisions, and the engineer who argues hard is frequently the reason a bad idea died. I can name several things that did not ship because one person would not let them through, and I was on the wrong side of at least two of those. Every mechanism in this piece can be turned into a device for making a room pleasant, and a manager who reads all disagreement as friction to be reduced will build a calm team that ships the wrong thing on schedule.

So the claim gets narrower. I am not arguing for less conflict, and if I had to pick between a team that argues too much and a team that agrees too easily I would take the first without hesitating, because its failure mode is visible and the other one is not. What I am arguing for is that arguments get settled by something checkable and then actually end. A team where disagreement is cheap will have more of it, which is the point, and will be noisier in the specific meetings where that matters.

I would rather not end this tidily, because it does not end tidily. Some of these situations do not resolve. Some people do not change, and are entitled not to; a person who is difficult in review and excellent at the work is not a defect to be corrected. The direct conversation sometimes goes well, sometimes changes nothing, and sometimes ends with the team being better off without somebody. That is a real loss of capability, and anyone who calls it removing a blocker has not counted properly.

What I hold onto is the thing that is cheap and available in any room. The price of disagreeing is being set right now, by whoever is hardest to disagree with, and it is being paid in objections nobody ever hears. Lowering it does not require anyone to be a different person. It requires the criteria on the table before the proposal is, the dissent to have somewhere to live, and somebody paying attention to who has stopped talking. And when the meeting finishes early, treating that as a question rather than a result.