The Lindy Effect: why the old survives – and when it's still right to bet on what hasn't yet

In the article on thinking in scenarios, we landed on the cobra effect – the clearest example of what happens when a decision isn't thought through to its consequences. This article is about a related but different question: not what happens when an incentive backfires, but how you judge what's worth trusting in the first place. The answer starts at a delicatessen in New York.

A deli, an observation, a law

In the 1960s, comedians at Lindy's in Manhattan used to debate whose careers would last. The observation was strange: a comedian's expected remaining time on stage didn't seem to shrink with age – it grew in proportion to how long they'd already been active. The longer the career behind them, the longer the career ahead.

Mathematician Benoit Mandelbrot later formalized the idea, and Nassim Taleb popularized it in Antifragile and Skin in the Game. Today it's known as the Lindy effect: for non-perishable phenomena – ideas, institutions, technologies, works of art – expected remaining lifespan increases in proportion to how long something has already existed.

It's the opposite of how biological creatures age. A 90-year-old has a shorter expected remaining lifespan than a 20-year-old. But a text that's been in print for a thousand years statistically has longer left than a book published last year. The wheel, scissors, and the knife have existed for millennia and will likely persist. Last year's hyped app is a far less certain bet.

Why the old wins

The logic is simple enough. Something that has survived a long time has already been tested against a long series of shocks, fads, and competing alternatives – and made it through. Survival itself is proof of robustness that no single analysis in the present moment can provide. The new, by definition, hasn't passed that test yet, however promising it looks right now.

It doesn't apply to everything old – only to what's information-based or cultural rather than physically perishable – and it's a probabilistic rule of thumb, not a guarantee. But as a starting point for judging what deserves your trust, it's unusually accurate.

The obvious objection

Here's the question that makes the Lindy effect genuinely interesting rather than just a curiosity: if the proven is what you should trust, why does every well-run technology organization disproportionately invest attention, development time, and capital in exactly what isn't proven yet? Why chase faster returns in the new instead of consistently weighting toward what has already proven it holds up?

The answer is that the Lindy effect says something about survival probability, not about return potential. Mature technology has already captured most of its economic value – it sits far out on its S-curve. New technology has a much higher probability of failing outright, but if it succeeds, the return can be many times larger, because it's still early on its own curve. That's not a contradiction of Lindy logic. It's a different question: not "what survives" but "where is the asymmetric upside."

The barbell: holding both truths at once

That's exactly the combination Taleb himself advocates in Antifragile, under the name of the barbell strategy. The idea isn't to distribute capital or trust evenly according to age and proven status. It's to deliberately polarize: the bulk of resources in the robust and proven, a small but real share in high-risk bets with capped downside and unlimited upside – never a large, untested commitment in the middle that's neither safe nor has the convex upside.

Capped downside means the maximum loss is known and limited before you commit – you can lose exactly what you staked, never more. A pilot budget with a firm ceiling, a small AI experiment confined to one process, an option-like engagement with a new vendor: if it fails, you lose the stake, and nothing else. That's the opposite of an open-ended downside, where a failure can spread uncontrolled – if your entire customer data flow gets rebuilt around a single, unproven vendor, and that vendor turns out unreliable, the loss isn't limited to what you paid for the service. It becomes whatever it costs to disentangle your core operations from a dependency you no longer control.

Applied to an organization's AI investments, the principle becomes concrete. Machine learning as a field has actually existed since the 1950s–60s – that gives the technology category itself some Lindy weight. But the specific model, vendor, or architecture you choose this year does not. The distinction matters for where a tool belongs in the portfolio: experiment broadly and disproportionately with the new, but avoid making a single, unproven vendor a sole critical dependency for core operations.

Delayed feedback: why capped downside matters more than it seems

There's a reason capped downside is harder to respect in practice than it sounds in theory, and it's about time rather than willpower. The field of system dynamics – Jay Forrester, later Donella Meadows and John Sterman – has shown for decades that most organizational mistakes aren't caused by bad decisions in the moment, but by delayed feedback: the gap between when a decision is made and when its consequences actually show up.

The longer that delay, the more an exposure can grow before anyone realizes it was wrong from the start. It's the same mechanism behind the bullwhip effect in a supply chain: small, seemingly reasonable decisions get amplified through several links before the deviation becomes visible further up, by which point it's already large. Applied to a technology investment: choose an unproven AI vendor as the core of a system, and the consequence rarely shows up immediately. It shows up once more processes have been built on top of the dependency, more teams have adapted their workflows around it, and more contracts have been signed assuming it will still be there – often eighteen months later, not eighteen days.

That's exactly why capped downside has to be decided before the bet is placed, not adjusted along the way "once we see how it goes." With a long delay between decision and consequence, you won't see how it's going until the exposure has already grown far past what you originally intended. A clear ceiling – a pilot budget, a time limit, an explicit criterion for when an experiment gets scaled up or shut down – is the only thing that actually keeps the downside capped through the whole period during which you can't see the result yet.

Where it actually goes wrong

The real violation of Lindy logic, then, isn't disproportionate attention on the new. It's putting capital without capped downside into it – building critical infrastructure on a technology as if it had already proven itself, in pursuit of faster ROI, and then trusting that the delay between decision and consequence will leave time to correct course later. That's when you swap a robust core for a fragile one, and delayed feedback means no one notices until it's already costly.

And as in Skin in the Game, it's rarely the person making that decision who bears the full downside if it goes wrong. The executive who wants to demonstrate rapid transformation by making a single, unproven AI vendor the core of a critical system rarely risks their own position as much as the organization risks its resilience. The question to ask before the decision, borrowed from The Black Swan: what do we lose, and who bears that loss, if this specific choice turns out to be wrong in two years?

It's the same kind of question the scenario series asks about the future at large – just turned inward, toward one's own investment portfolio, rather than outward toward which scenario actually unfolds. The next article in the series on consequence-thinking takes that step: what happens when several such decisions, each reasonable on its own, reinforce each other across an organization – and why that effect rarely shows up until it's already expensive.