An Experiment to Live Inside

Yesterday, Associated Press reported that Warren Buffett stepped down as chairman of Berkshire Hathaway, a role he had held since 1970. The 96-year-old remains a director and becomes chairman emeritus; the chairmanship passes to his son, Howard Buffett, who’s been on the board since 1993.¹ Operational control already changed when Greg Abel became CEO at the start of this year. What ended yesterday is more symbolic: the last formal thread tying the company to the man who built it. Buffett’s own description of the arrangement: »Think of Howard as a policy the shareholders own and hope never to claim against«.²

»That explains how the rule survives changing markets; it doesn’t explain whether it survives changing hands«

Some people build a house. Buffett built an experiment to live inside for sixty years.

It’s hard to overstate how much the environment changed during that experiment. Try turning it around: describe 2026 to the man who gave himself the job in 1970, when there was no Internet, no mobile phone, not even a personal computer. Tell him he’ll live through a dozen market crashes and a financial crisis nobody saw coming, that an actor and a TV star will become presidents of the United States, that hijacked airliners will bring down Manhattan’s tallest towers, that money will travel as numbers nobody can hold in their hand, and that machines will write, paint, and argue back before breakfast. He would have called you a lunatic. And yet the basic machinery of Buffett’s thinking during those delirious years barely moved.

That machinery is worth unpacking a little: buy businesses you understand, judge them by their underlying economics rather than their share price, prefer durable advantages and trustworthy managers, think like an owner rather than a trader, know your circle of competence, be willing to wait.³ These are almost comically underspecified instructions: more »Yoda« than »manual«. They tell you how to think without telling you what to think, and that is precisely their usefulness.

A rule that says »buy companies in industry X« is vulnerable to history. A rule that says »buy companies with durable economic advantages« leaves the future open: it specifies the desired behaviour without specifying what the environment must look like when that behaviour is applied. That’s part of the answer. But it isn’t all of it.

Much of what we build, in finance as well as in artificial intelligence, tries to predict what happens next: the market going up or down, the next word, the protein structure, the classification. But Buffett realised there’s another way to build an intelligent system, without needing to know what happens next. And he constructed it in a way that actually seems strangely aligned with how we’re now learning to build AI: make the thing, run it, find out what it can do and where it breaks, then revise it. Discovery instead of specification. The difference is that Buffett’s version was making money from day one.

This is what makes Buffett’s non-prediction principle a peculiar form of resilience. A system that depends on an accurate forecast is only as resilient as that forecast. A system built to function despite the inevitable failure of trying to predict the future has a different relationship with uncertainty: it doesn’t solve it, but makes it less consequential. That explains how the rule survives changing markets; it doesn’t explain whether it survives changing hands.

Succession is where that question gets tested. Berkshire can write down its rules and teach them to new managers, which is more or less what happened at the top level yesterday. But rather than the rules themselves, the harder thing to hand over is the calibration process. »Know your circle of competence« sounds simple until something appears that nobody’s seen before: is it outside the circle, or does the circle need to expand? Maybe the real legacy isn’t so much the principles but the conversion of judgement into rules that other people can apply – and the harder question of whether they can keep revising them.

There’s an irony here: prediction built the bridges, priced the insurance, trained the machines. There’s nothing inherently superior about refusing to predict. What matters is narrower: not whether a system uses prediction, but whether it depends on prediction to survive. We can predict that the world will change. We usually can’t predict how.

The better system – at least when it comes to making enormous amounts of money – simply doesn’t need to.

So what actually makes a rule survive massive changes to its environment? Not its wording, but a willingness to keep testing that wording against the world and revising it when it breaks: discovery, not specification, run for sixty years on nothing but discipline. Buffett didn’t just write the experiment’s rules; he was the one who kept living inside it, testing every day whether it still held.

Berkshire is about to find out if someone else can move in.

References