What ecology already knows about a software industry that stopped changing
Almost every banana sold in Europe is the same plant. Not the same species - the same individual, cloned by cutting for a century. It is called the Cavendish, and it was chosen in the 1950s because the previous banana, the Gros Michel, had just been wiped out by a soil fungus it had no defence against. The Cavendish was resistant. It was also, being a clone, incapable of ever becoming more resistant. A new strain of the same fungus is now moving through plantations on three continents, and the industry’s plan is roughly what it was last time: find a survivor, clone it, hope.
Nothing about this is a banana problem. It is what happens to any population when you remove variation and optimise what is left. Ecology has spent a hundred years working out the rules, and the software industry is currently breaking most of them.
Rule one: no variance, no adaptation
In 1930 R.A. Fisher wrote down what is still called the fundamental theorem of natural selection. In one sentence: the rate at which a population can improve its fitness is equal to the amount of genetic variance in that population. Not the average quality. The spread. A population of excellent identical organisms cannot adapt to anything, because selection has nothing to choose between.
Look at how software gets funded and ask what it selects for. A venture investor underwrites pattern. The pitch that clears the process is the one with a comparable: the Uber of X, the Stripe for Y, a known model with one variable changed. Anything without a comparable has no market-size slide, and anything without a market-size slide cannot be priced, and anything that cannot be priced does not get money. This is not stupidity. It is a rational method for reducing risk. It is also, precisely, the selection of clones.
The result is an industry with a great deal of quality and almost no variance. Fifty companies doing the same thing with different logos is the Cavendish plantation: uniform, productive, and defenceless against the one thing none of them has seen.
Rule two: the fire you prevent is the fire you get
For most of the twentieth century, forest agencies in the American West put out every fire they could find. It worked. Fires got rarer. It also meant that the small, frequent burns that used to clear undergrowth every few years stopped happening, and the undergrowth accumulated for decades. Foresters have a term for it: fuel load. When a fire finally came that could not be put out, it had fifty years of material to burn through, and it burned the canopy instead of the floor.
Small failures are how a system learns what does not work. The software industry has spent fifteen years making sure they do not happen. Companies that should have died were kept alive by capital that cost nothing to borrow. Products that would have failed were bought before failing and folded into something that could absorb the loss. Every one of those prevented failures was information the ecosystem never received, and every one of them added to the fuel load: the accumulated mass of things that exist because nothing was allowed to test whether they should.
A market in which nothing small ever burns is not a healthy market. It is a forest waiting for a summer.
Rule three: the Red Queen only runs when someone is chasing
In 1973 the evolutionary biologist Leigh Van Valen noticed something strange in the fossil record: the probability of a lineage going extinct did not decrease with how long it had already survived. Being well adapted bought you nothing, because everything around you was adapting too. He named it after the Red Queen in Through the Looking-Glass, who explains that here you must run as fast as you can just to stay in place.
That running is where new capability comes from. Predators get faster because prey got faster; prey got faster because predators did. Remove one side and the other stops. It has no reason to continue.
This is what an acquisition does to a category. When the leading company buys the competitor instead of outrunning it, the co-evolution ends. The feature does not get better; it gets bundled, and then it gets deprecated. The pattern is now regular enough to have a shape: a small product does something the platform does not, gains users, gets acquired, and is shut down within a year or two. The users are absorbed. The capability is not. Anyone who watched Skype, Wickr or Skiff go through that cycle has seen a species removed from an ecosystem so that the species eating it would not have to change.
Rule four: generation time
An elephant and a bacterium are both very well adapted to what they do. Only one of them can respond to a new antibiotic by Thursday. The difference is generation time. Evolution happens between generations, and a lineage that reproduces every twenty minutes will always out-adapt one that reproduces every twenty years, regardless of which organism is more impressive to look at.
The largest software companies have the longest generation times in the industry. A change of direction requires committees, reorganisations, quarters. Their advantage is scale; their cost is that scale is slow, and slow is the one thing you cannot buy your way out of when the environment changes.
Bacteria have a second trick worth noticing. They do not only inherit genes vertically from their parent; they swap them sideways with whatever is nearby. Horizontal gene transfer is why antibiotic resistance spreads across species that have never met. Software has the same mechanism, and it is called open source. A closed codebase inherits only from itself and, like any asexual lineage, accumulates what geneticists call Muller’s ratchet: errors that can never be edited out because there is nothing to recombine with. An open one borrows from everything. Over a long enough period the difference is not in features. It is in whether the organism is still alive.
Rule five: stasis is not stability
Palaeontologists spent a long time being puzzled that the fossil record does not show gradual change. It shows long periods in which nothing happens, punctuated by short intervals in which everything does. Eldredge and Gould called it punctuated equilibrium in 1972, and it reframed what a period of stasis means. It is not evidence that a system has reached its final form. It is evidence that the pressure has nowhere to go yet.
The software stack has been in stasis for roughly a decade. The same few platforms, the same revenue model, the same monocultures of framework and cloud and metric. It looks like maturity. Ecologically it looks like the late Cretaceous: very large, very specialised, very well suited to a world that has stopped needing to be anything else.
What ends such periods is never internal. The system does not reform itself, because it is optimised for the conditions that made it, and reform would cost fitness under those conditions. It ends when something arrives that the incumbents were never selected to handle. And what survives it is, reliably, not the largest or the best adapted. After every mass extinction the survivors are small, generalist, and cheap to run. Palaeontologists call it the Lilliput effect. The mammals that inherited the world were not better dinosaurs. They were the things living under the dinosaurs’ feet that could eat almost anything and did not need much of it.
The apex strategy
Nobody has made the stagnation argument louder, longer or with more money behind it than Peter Thiel. The Founders Fund manifesto put it in one line - we wanted flying cars and got 140 characters - and Zero to One turned it into a model. Progress is either horizontal, going from 1 to n by copying something that works, or vertical, going from 0 to 1 by making something that did not exist. The industry, he argued, does almost nothing but the first. Underneath sits René Girard: people do not want things, they want what other people want, and competition is not a search for value but a crowd converging on the same prize. Read that against Fisher’s theorem and it is the same diagnosis in different vocabulary. Copying is the removal of variance. Mimetic desire is how a population becomes a clone.
So far the model and the fossil record agree. The disagreement is over the cure.
Thiel’s prescription is escape. Competition is for losers; the goal is a monopoly, because only a monopoly has the margin to think about anything other than survival. Find a small market, own it completely, expand from there; be the last mover, not the first. And fund accordingly: returns follow a power law, so back the single outlier and treat the rest as noise. Ask every founder what important truth very few people agree with them on, and fund the ones with an answer.
In ecological terms this is the strategy of the apex organism, and it has a known outcome. The species that escapes the Red Queen does not use its freedom to explore. It specialises, grows, and lengthens its generation time, because under no selection pressure there is nothing to punish that. A monopoly is a climax community with a share price. The model that correctly diagnoses the monoculture prescribes founding one.
What the founders who took it literally will accomplish is not nothing, and it would be dishonest to pretend otherwise. The model is right that copying is worthless, and the people who believed that built a handful of real 0-to-1 outcomes in domains the rest of the industry had written off as impossible - payments once, rockets, and the software that now runs inside intelligence agencies and militaries. That is more than the 1-to-n crowd will ever show.
But watch where the escape from competition leads once it is complete. The most durable moat is not a product. It is a customer that cannot leave, and no customer is harder to leave than the state. Palantir, Anduril and the companies now forming around them have escaped the market by becoming part of the environment: the clearance, the contract, the integration so deep that the government’s own processes run through the vendor. Biologists call this niche construction - beavers, corals, the organism that stops adapting to its surroundings by rebuilding the surroundings around itself. It is a spectacular strategy right up to the moment the environment changes for reasons the organism does not control, at which point the most specialised thing in the system is the first to go.
There is a second irony the school should have seen coming, because its own philosopher wrote it down. Ask a whole generation of founders the contrarian question and they will give you the same answer. Contrarianism with a template is mimesis, and a secret that clears a funding committee is not a secret. The model does not select for the truly new. It selects for the fundable new, which is the familiar wearing a black turtleneck.
Put the pieces together and the accomplishment comes into focus. A few very large, very definite organisms, welded to state budgets, ending the Red Queen in their domains, with generation times as long as any incumbent’s but disguised by founder charisma. In his recent lectures Thiel casts a single global authority as the thing to fear most; the model, followed to its end, produces the same centralisation on a smaller map - sovereignty concentrated in a handful of private apex firms rather than in one public one. That is not a renaissance. It is a more durable stasis than the one it replaced, and a heavier fuel load. The late Cretaceous, with a defence budget.
And the Lilliput effect does not read the pitch deck. When the fire comes, it will treat the apex firms exactly as it treats the platforms they displaced.