Camels

There’s an old engineering joke that a camel is a horse designed by a committee. It’s unfair to camels, which are superbly adapted animals, but everyone knows what it means the first time they hear it. Someone wanted a bigger water tank. Someone wanted it to survive the desert. Someone wanted to cut costs on the legs. Everybody got a little of what they wanted, and nobody got a horse.

I want to defend a stronger claim than the joke. The joke says committees produce ugly results. I think they produce worse results: not less elegant, but actually inferior to either of the positions that went into the room. And the reason is structural, not aesthetic, which means it shows up everywhere. In products, in teams, and most expensively, in governments.

Compromise feels like the adult thing to do. You give a little, I give a little, and we meet in the middle like reasonable people. The whole moral weight of the word is on your side. Nobody writes essays titled “in praise of refusing to budge.” But I think the intuition is wrong, and I think I can show you why with a picture.

Imagine you’re trying to find the highest point in a landscape. You and I each have a candidate. I’m standing on one hilltop, you’re standing on another. Neither of us can see a higher peak, so each of us is, in the language of optimization, at a local maximum. We’re both, in our own way, right.

Now we compromise. We agree to meet in the middle, halfway between my hill and yours. Where do we end up? In the valley. The midpoint between two hilltops is almost never higher than either hilltop. It is, very often, lower than both.

That’s the whole problem with compromise in one image. When the landscape is smooth and bowl-shaped - what mathematicians call convex - averaging two decent answers gives you a better one, and compromise is genuinely wise. But almost nothing interesting is convex. Real problems have multiple peaks separated by valleys, and on that kind of terrain the average of two good solutions is reliably worse than either.

This isn’t a metaphor I’m stretching for effect. It’s the literal reason you can’t train a neural network, design a wing, or route a chip by averaging two good configurations. Take two trained networks that each work, average their weights, and you usually get one that works worse than both, because the line connecting two good solutions runs straight through the territory of bad ones. Engineers know this in their bones. They just don’t usually notice that the politics down the hall obeys the same math.

There’s a second thing the camel joke gets right that the optimization picture alone misses. A good solution isn’t a bag of features you can mix and match. It’s a set of choices that fit together. The horse’s legs make sense given its weight. Its stomach makes sense given its diet. Its speed makes sense given what’s chasing it. Pull one piece toward the camel and the rest stops cohering.

A plan is a chain of dependent bets. “Build coal plants” is a coherent plan: you commit to the emissions, you get the cheap baseload power, you design the grid around it. “Build renewables and decommission everything else” is also a coherent plan: a different bet, a different grid, but internally consistent. The compromise - keep half the coal, half-build the renewables, decommission on a timeline nobody believes - is not a third plan. It’s two plans sawn in half and nailed together, and the seam is where everything fails. You pay the emissions and the capital cost and the reliability risk, and you get a grid optimized for neither world.

I wrote about political velocity, why two-party systems can change direction and coalition systems mostly can’t. The argument there was about speed. This is the argument underneath it. A three-party coalition can only pass what lies in the overlap of what all three can stomach, and that overlap is never a plan anyone designed. It is the intersection of three plans, which is to say, the valley between three hills. The Greens didn’t run on building coal plants. The liberals didn’t run on expanding the welfare state. So the country does a little of each, half-heartedly, and calls the camel a summer of reform.

The immediate objection to all this is: so what do you want instead, one person ramming through whatever they like? No. The alternative to compromise is not the absence of disagreement. It’s putting the disagreement in the right place in time.

Andy Grove used a phrase at Intel that Bezos later made famous at Amazon: disagree and commit. The Amazon version is blunt about it. Leaders are obligated to challenge a decision when they think it’s wrong, even when it’s exhausting. Then, once the decision is made, they commit to it completely. The official wording includes a line I find clarifying: they do not compromise for the sake of social cohesion.

That last bit is the whole thing. Compromise is what you reach for when the goal has quietly shifted from “make the best decision” to “make everyone feel heard.” Those are different objectives, and the second one is the enemy of the first. Disagree-and-commit separates them. You argue first, really argue, with backbone, trying to win. Then someone decides, and everyone rows in the same direction as if they’d agreed from the start. The dissent happens before the commitment, at full volume, and then it stops.

Notice what this gives you that compromise cannot. You get a coherent decision - one of the hilltops, chosen on purpose, rather than the valley nobody picked. You get speed, because the argument has an end. And you get someone who owns the outcome. A compromise has no author, which is why a compromise can never be wrong: there is always a partner to blame. A committed decision has an author, which means it can be evaluated, and being able to be evaluated is the precondition for ever getting better.

This is the part that sounds insane until you sit with it. A coherent decision you later discover is wrong is more valuable than a compromise that is vaguely, untestably okay. The wrong answer teaches you something. The mush teaches you nothing.

When you build the coal plant and it turns out to be a mistake, you learn precisely that. The bet was legible, the result was legible, and next time you bet differently. You can U-turn. When you build the half-coal-half-renewable camel and the country muddles along underpowered and over-budget, what did you learn? Nothing, because there was no clean hypothesis. You can’t iterate on mush. There’s no edge to push against, no clear thing that failed, no one who owns it. Compromise doesn’t just produce a worse answer. It produces an unlearnable one.

This is why the country that can be wrong fast beats the country that is slowly, consensually almost-right. Wrongness is recoverable. It’s a position in the space you can move away from. The camel is not a position. It’s a smear across the space, and you can’t move away from a smear because you were never anywhere to begin with.

I should be honest about where this breaks, the same way I tried to be in the velocity essay, because a one-sided argument is itself a kind of camel.

There is a real version of meeting in the middle that isn’t compromise at all, and it’s worth naming so we don’t confuse the two. When two positions differ by degree rather than by kind - how high to set a tax, how many weeks of leave, where to draw a line that has to go somewhere - the landscape between them really is convex, and the midpoint can be a sensible point. Splitting the difference there isn’t mush. It’s arithmetic. Don’t average two architectures. Do average two thermostat settings.

But here’s where I part company with most people, including the ones who think they agree with me. The usual move is to grant that compromise is bad in engineering and politics, where things are technical, and then exempt the one place it supposedly matters most: personal relationships. Compromise, we’re told, is the very substance of love. A good marriage is a long series of meetings in the middle.

This is exactly backwards. Relationships are where the valley is deepest.

Watch what actually happens when two people split the difference. One wants a child, the other doesn’t. One wants the city, the other the countryside. The compromise positions - have the kid but resent it, move to the suburb neither of you chose - are the valley, and everyone can feel it. Esther Perel has a name for the trap that leads there. She calls it splitting the ambivalence. When a couple polarizes into “I want it” and “I don’t,” each person has quietly taken up one half of a question that lives, whole, inside both of them. The one who says no has longings too. The one who says yes has doubts too. They have outsourced their own internal complexity to each other and turned it into a tug of war. The midpoint of a tug of war is just the spot where the rope stops moving. Nobody chose it. It’s the average of two amputated positions, which is to say, a camel.

The escape is the same one disagree-and-commit describes in a company. Not the middle. A third hilltop. You stay in the argument long enough to find the option that meets both people’s actual needs, the one neither of you could see while you were each defending half of yourselves. This only works if the disagree half is real. Skip the listening and what you have isn’t conviction, it’s stubbornness in a costume. The real version is slower than splitting the difference and far more uncomfortable, because it requires admitting you wanted contradictory things all along. But it produces something coherent: a decision with an author, a life you can stand behind, a thing you can later discover was wrong and recover from. The compromise produces only the slow leak, the resentment that has no event to point to, because nothing was ever decided, only conceded. Compromise is splitting the difference. What saves a marriage is refusing to.

So the harder claim stands. Compromise isn’t the price of love. It’s the thing that quietly empties it, one conceded Sunday at a time. The two cases where meeting in the middle is fine - a difference of degree, and committing to a partner’s decision you argued against and lost - aren’t compromises either. The first is arithmetic. The second is disagree-and-commit wearing a wedding ring.

Compromise survives as a virtue mostly because it’s polite. It lets everyone leave the table - the negotiating table, the cabinet table, the kitchen table - feeling reasonable. But the politeness is paid for by the result, and the bill is large. You trade a coherent answer for an incoherent one, an author for an alibi, a thing you can learn from for a thing you cannot.

The horse was never going to make everyone at the table happy. That was the point of the horse.