expectedwrong hindsight

The Tractor Problem

Why the thing that does everything well enough beats the thing that does one thing perfectly.

2 min read 388 words #economics #strategy #generalization #tractors #geopolitics
hindsight — nailed it

The tractor metaphor was the whole thesis of the next three years. LLMs are the tractor. 75%% at everything beats 99%% at one thing, because you only buy one and it covers the farm. Every company that bet on narrow specialist AI got outrun by general-purpose models doing everything at B+ level. The math was brutal once you spelled it out, and it stayed brutal.

On the island, the logic goes like this: you want to blow snow, you get a snow blower. You want to mow grass, you get a zero-turn. You want to dig, you get an excavator. You want to move weight, you get a skid-steer with pallet forks. Each machine does its one job at something like 99% effectiveness.

The tractor does all of it at 75%.

And yet — you get a tractor.

The math is brutal once you spell it out. Fifty specialized machines to cover every task at 99%. One tractor to cover all of them at 75%. The question isn't which approach is better in some abstract sense. The question is what you're actually optimizing for, and whether you've been honest with yourself about the answer.

Most people need coverage. They tell themselves they need excellence.

This is not really about tractors.

Bangladesh decided — actually decided, as a matter of national economic policy — that the entire country would become a garment factory. Not metaphorically. They aligned their labor force, their infrastructure, their trade relationships, around one thing. And it worked, in the specific sense that Bangladesh is now extremely good at making clothes and has built real GDP on the back of it.

India, meanwhile, is a tractor. Huge, capable across an extraordinary range of industries and skills and outputs, and perpetually constrained by the coordination costs of being everything to everyone. The Daily had an episode on this recently — the argument being that India's generalization isn't just an economic profile, it's a source of internal competing interests that make it structurally difficult to commit to anything the way Bangladesh committed to garments.

The tractor can't fully optimize any single attachment because it has to be compatible with all of them.

The thing nobody says out loud: the specialized machine is only a good bet if you're right about which task matters. Bangladesh bet on garments. That bet has a geopolitical half-life. Supply chains move, labor costs rise, synthetic materials eat into traditional textile demand — you can imagine a dozen ways that being the world's shirt factory becomes a worse position to be in.

The tractor survives because it never concentrated its exposure.

75% across everything is a hedge. It feels like underperformance. It might be the whole game.