Superlinear Returns: Why the World Doesn't Pay You What You Put In

"You get out what you put in." Every teacher and coach said some version of it, and they meant well. As a description of how the world actually rewards performance, it's almost never true.

If your product is half as good as your competitor's, you don't get half the customers. You get roughly none, and you go out of business. Returns for performance aren't linear. They're superlinear — and once you see this, you can't unsee it in business, fame, power, knowledge, or benefit to humanity. Everywhere, the rich get richer. This isn't a flaw in capitalism you could rule-change away. It's a feature of the world, and if you're ambitious it's the wave to surf rather than the current to fight.

Two engines, and only two

However varied superlinear returns look, they trace back to two causes: exponential growth and thresholds. Learn to spot these two shapes and you have a lens for reading almost any competitive situation.

Exponential growth happens whenever how well you do next depends on how well you've done so far. Startups grow this way; so do bacterial cultures, audiences, and reputations. The startling formalization: if your growth rate is proportional to your performance, then reward for performance p over time t scales as p^t. Performance sits in the exponent. A small edge, compounded over enough cycles, becomes an unbridgeable gap. That's the whole game.

This feels alien because almost nothing in human history was exponential, so we built no intuitions for it. The tell is the courtier who asks a king for one grain of rice doubled each day, and bankrupts the kingdom by month's end. Every child is surprised, because our instincts are stubbornly linear.

Thresholds are the other engine: "winner take all." Win a match by an inch or a mile, you get exactly one win. Note the mechanism carefully, because most people misattribute it: the step function isn't created by competition, it's created by a threshold in the outcome. Proving a theorem, hitting a target, learning to ride a bike — below the line, nothing; above it, everything. You can have thresholds with no rival in sight.

The multiplier: the two engines feed each other

The load-bearing part of the model is that the engines compound each other. Crossing a threshold triggers exponential growth: the winning side in a battle takes less damage, which makes it likelier to win the next. And exponential growth lets you cross thresholds: in a market with network effects, growing fast enough locks competitors out entirely.

Fame runs on both — fans recruit fans (exponential), but there's finite room on the A-list in any head (threshold). Learning runs on both — knowledge compounds, and some thresholds, like literacy, are machine tools that make acquiring everything else faster afterward. This is why the biggest returns come from discoveries: knowledge is fractal, and pushing hard at the edge of one field occasionally drops you into a new one where you get first crack at everything inside.

The rule, and the test that keeps you honest

The portable rule: seek work that compounds. It compounds two ways.

Directly — infrastructure, an audience, a brand, a codebase — where each cycle's output is next cycle's input. Ask: does this round's success make the next round easier, or do I restart from zero each time? Waiting tables, most consulting hours, most content-for-hire: you sell the same hour repeatedly and it never gets cheaper to produce. Linear. Writing that accrues an audience, code that becomes a platform, skills that stack: superlinear.

Through learning — and here the model does its most useful work, because this route can look exactly like failure while it's happening. You can miss your immediate goal entirely and still be compounding — in yourself. This is why Silicon Valley funds people who've flamed out. The company didn't compound; the founder did.

But "always be learning" is where most people lie to themselves, so it needs a real test, not a slogan. After a loss, comforting yourself with "I learned a lot" is free and usually false. Distinguish the two like this:

• Compounding knowledge produces a shorter path next time. You can name a specific thing you will now do differently, and it generalizes beyond this one situation. Before, you'd have made the mistake again; now you'd catch it. That's exponential growth wearing the disguise of failure.
• Accumulating losses produces a story. You feel wiser but can't state a decision you'd change, or the "lesson" is so specific it never recurs, or so vague ("work harder," "trust the process") it's useless. That's a linear treadmill wearing the costume of effort.

The concrete drill: after any setback, before you're allowed to say you learned, write one sentence in the form "Next time in situation X, I will do Y instead of Z." If you can't, you didn't learn — you rationalized. If you can, and X is broad, you just got exponential growth for the price of a linear-looking failure.

Point both questions at your own week — Does anything I'm doing make the next round cheaper? When something failed, can I name the sentence? — and you'll stop measuring your progress by the flat numbers your teachers taught you to watch, and start measuring the slope.

Distilled from Paul Graham Essays