The Same Two Animals

Let me tell you about two companies, and you tell me which decade we are in.

The first company wants to be everywhere. It courts the largest incumbent in its industry and rides that partnership to ubiquity. It opens as many channels as it can, ships fast, and measures itself in users. Its great talent is making itself difficult to avoid. Its founder is a fluent, slightly awkward evangelist who talks about the technology as a force of history.

The second company wants you to care whose product you are using. Its audience is smaller, its business is more selective, and it says no more often. It guards its crown jewels from rivals. It publishes documents about how its product is supposed to behave, and treats that as part of the product. It wins the professionals who make things, and the professionals love it in a way nobody loves the first company. It buys expensive television time during the Super Bowl to mock the incumbent’s way of doing things. Its executives get accused of wanting to control what people do with the technology. Its admirers call this taste. Its critics call it control.

You already know that both descriptions fit 1985 and both fit 2026. The map is clean. In the 1985 story, the first company is Microsoft and the second is Apple. IBM is the incumbent Microsoft courted. In the 2026 story, the first company is OpenAI and the second is Anthropic. Microsoft is the incumbent OpenAI courted. NeXT belongs inside Apple’s story: it is the breakaway company Jobs founded after leaving Apple, much as Anthropic began with people who left OpenAI. The details are not interchangeable. Gates stayed in command of Microsoft, while Jobs was pushed out of Apple by its board in 1985; the board drama belongs to Altman in the newer story.

Apple’s “1984” commercial cast the Macintosh as liberation from a grey IBM world. Anthropic’s four-ad Super Bowl campaign in February 2026 mocked a chatbot that interrupts a heart-to-heart to recommend insoles for short kings, and Sam Altman called the ads “so clearly dishonest” and accused Anthropic of wanting “to control what people do with AI.” Altman was removed by OpenAI’s board on Friday, November 17, 2023, reached an agreement to return the following Wednesday, and formally returned the next week. Different incumbent, different grievance, the same invitation: you are the sort of person who notices what is wrong with the default.

When I first laid the two stories side by side I did what everyone does: I laughed, and then I got a little uneasy. The match is too good. It is good in the way a forged painting is good, where the brushstrokes line up so well you suspect someone was tracing.

It is tempting to keep matching details until you have a magnificent historical crossword. But IBM occupies the incumbent slot beside Microsoft, and NeXT is the breakaway company inside Apple’s story; neither replaces one of the two animals. The resemblance that survives without that confusion is simpler and more interesting. Once again, a technology that promises to change everything has grown two competing temperaments: be everywhere, or be unmistakably yourself.

We usually explain the temperaments by pointing at the people on top. Gates was Gates. Jobs was Jobs. Altman and Amodei have their own convictions. All true. But why do different people, four decades apart, confronting a technology that shares nothing at the level of physics with the last one, end up facing such familiar choices?

Perhaps the habitat has something to do with the animals.

Marsupial wolves

In 1936 the last thylacine died in a zoo in Hobart. If you look at a photograph you see a wolf: long skull, dog-like teeth, the lean body of a predator. Except it was a marsupial. It carried its young in a pouch. Its last common ancestor with the actual wolf lived something like 160 million years ago, back when both lineages were small furry things scuttling under dinosaurs. The thylacine and the wolf did not copy each other. They could not have; an ocean and an eon separated them. They arrived at a similar outward form because the niche “medium-sized mammal that eats other mammals” has, it turns out, a limited set of good answers.

The resemblance was also incomplete, and that matters. Earlier studies of thylacine bones suggested a hunting style closer to ambush than to a wolf’s long pursuit. A newer study finds a distinctive high-impact snapping bite and argues that its cranial function cannot simply be inferred from living mammals. Evolution produced a look-alike, not a duplicate. Keep that in mind whenever someone starts comparing companies to animals, including me.

Biologists call the independent evolution of similar features convergence, and the textbooks are full of it. Dolphins and ichthyosaurs. Cactus in the Americas and euphorbias in Africa, unrelated plants that look remarkably alike because the desert asks the same question of both. Complex eyes evolved independently many times. The lesson is not that life is unimaginative. The lesson is that the environment is a sculptor with strong opinions, and organisms are the clay.

Now hold that thought and ask: what happens when two lineages occupy the same niche at the same time?

The Russian ecologist Georgy Gause worked on this in the 1930s with jars of paramecia. Two species competing for exactly the same resources in exactly the same way do not share the jar; one drives the other out. That is the competitive exclusion principle, and it has an escape clause that matters more than the rule. The two can coexist if they stop being the same. If one shifts toward slightly larger prey and the other toward slightly smaller, the pressure on each is relieved. Darwin’s finches do this with beak size: where two similar species share an island, their beaks diverge more than either would on its own. Biologists call it character displacement, and it means something wonderful. The presence of a rival is itself a sculptor. Your competitor shapes you into what it is not.

Nothing here guarantees a tidy world of pairs. One competitor can simply vanish. Several can find several different livings. Displacement is a tendency, not a law. But it is a strong tendency, and it gives us two forces to hold in mind at once. Convergence pulls each organism toward the shape its niche demands. Displacement pushes two rivals in a shared niche apart from each other. Run both on a pair of surviving species and a pair of complementary shapes can emerge, with the shapes depending on the niche more than on the species.

I think this is what we are looking at. The personal computer and the language model are, from the market’s point of view, the same niche: a general-purpose technology, useful to everyone, sold through both the mass channel and the professional channel, where the customer cannot evaluate the product on specifications and has to pick a temperament instead. That niche has been occupied twice now. Both times it sculpted the same two animals.

The ubiquity animal and the integrity animal

Call the first shape the ubiquity animal. Its strategy is to be the substrate everyone else builds on: get onto every desk, into every workflow, be the default. Microsoft’s pivotal move was to license DOS to IBM while keeping the right to license it to every clone-maker on earth, so that a customer could buy a machine from anyone and still become a Microsoft customer. By the late 1990s something like nine out of ten personal computers ran Windows. OpenAI’s defining consumer move was ChatGPT, which by 2026 had more than a billion weekly active users and had begun putting advertisements on its free and low-cost tiers.

This kind of business learns to accommodate. Hardware varies. Customers have awkward old software. Partners want exceptions. Being available everywhere requires people who can negotiate, integrate, and keep things working despite the mess, and each successful accommodation makes the next one easier to justify. There is an enormous public good in this. A useful technology reaches people who cannot afford perfection or do not share the designer’s preferences. There is also a temptation: once you are underneath everything, you can make leaving expensive.

Call the second shape the integrity animal, using integrity in its older sense of wholeness. This animal wants its choices to hold together, from the metal to the pixel, and it sells the coherence of that control as a premium. It asks the customer to trust that there is a reason the pieces fit as they do, including the pieces the customer would rather change.

It says no a lot. In the older story, it says no to licensing its operating system to clone-makers, which Gates personally urged Apple to do in a 1985 memo to John Sculley, and which Apple refused. In the newer story, Anthropic says no to letting a rival use its software to build a competitor: in August 2025 it revoked OpenAI’s API access after learning that OpenAI engineers were using Claude Code just before GPT-5 shipped. It says no to advertisements; in February 2026 Anthropic made keeping ads out of Claude part of its identity, and then bought millions of dollars of television time to explain why ads did not belong in its product. There is something wonderfully Apple-like about that. It says no, in Anthropic’s case, to the Pentagon, which wanted access for all lawful purposes and got instead a company insisting on written limits around autonomous weapons and domestic surveillance, and which then designated Anthropic a supply-chain risk. The point is not who was right. The point is which animal picks that fight. The ubiquity animal does not pick that fight. The ubiquity animal signs.

The integrity animal also has taste, and it cannot help it. Look at the Claude interface with its serif type and warm terracotta, then look at the Mac with its Chicago font and its smiling icon, then look at ChatGPT and Windows, which are both perfectly fine and both look like something a committee approved. The integrity animal publishes a constitution for its model the way Apple published Human Interface Guidelines, a document telling everyone how the thing is supposed to behave. And it collects the same worshippers. In the 1990s the Mac’s true believers were the people who made things: designers, musicians, layout artists, a small cult that punched far above its market share. Today the people who make things are programmers, and programmers have adopted Claude the way designers once adopted the Mac. Same cult, new trade.

None of this makes the integrity animal virtuous by definition. Coherence can become rigidity. Trust can become a premium charged to people who have grown afraid to leave. For frustrated Mac users in 1995, the restrictions were evidence that somebody in California had confused their own preferences with universal law, and Claude’s refusals produce the same complaint today. Both animals can trap their customers. They offer different reasons to walk in.

How a strategy becomes a temperament

It is one thing to say a niche rewards two strategies. It is another to explain how a strategy gets into the bones of an organization, so that it persists after the founder stops attending every meeting. So picture the meeting.

Somebody proposes a lucrative partnership that requires an exception to a product rule. Half the room sees an opportunity. The other half sees the beginning of a slow surrender. The decision matters far beyond that contract. If the exception is made and it succeeds, the person who argued for it gains influence. The company hires more people who can deliver similar deals. Its customers come to expect flexibility. Within a few years a proposal to refuse a deal sounds less like principle and more like an inability to understand the business.

Now imagine the other decision. The company refuses, explains why, and wins customers who appreciate the refusal. Those customers pay for the promise and complain loudly when it is broken. Employees who take pride in protecting it stay. Applicants who admire it join. The next exception faces a room increasingly inclined to resist it.

Repeat this for a few years and you no longer need a founder in every meeting. The organization has developed preferences of its own. A successful choice attracts the customers, employees, and investments that make similar choices more likely, and the company gradually acquires a body suited to the way it earns its living.

The rival speeds this up. Once one company offers enormous reach, a challenger has more to gain from being distinctly worth choosing. Once the challenger becomes known for particular commitments, the larger competitor can appeal to everyone who finds those commitments inconvenient. Each makes the other’s alternative easier to understand. That is character displacement, conducted through hiring plans and pricing pages instead of beaks.

Sometimes the sorting is literal. Apple’s second act began when its founder was pushed out and went off to build NeXT out of spite and principle. Anthropic began in early 2021 after Dario and Daniela Amodei and a group of colleagues left OpenAI in late 2020, believing that OpenAI was commercializing faster than its safety work could follow. They left to do it the way they thought it should be done. The integrity animal is quite literally made of people who found the ubiquity animal’s temperament unbearable. You do not need to imagine anyone cynically inventing a principle for marketing. Convictions determine who leaves, who joins, and what the new company becomes. A person can believe every word of a mission statement while the market decides how much room that mission gets to occupy. Sincerity and selection work together.

Where the tracing goes wrong

If the story stopped there it would be a cute analogy, and cute analogies are cheap. What convinced me this is mechanism rather than coincidence is the places where the match breaks, because they break in exactly the ways an ecologist would predict.

The customers are swapped. In the mature version of the original market, Microsoft owned the office and Apple owned the living room and the art studio. Today the integrity animal is the enterprise-facing company: Anthropic says it is focused on businesses and developers, and surveys of enterprise spending on language models put it ahead of OpenAI. The ubiquity animal has its largest public reach through consumers. That looks like an inversion, until you notice what displacement actually promises. It does not promise that the same customers will be partitioned the same way. It promises that whatever axis is available, the two rivals will divide along it. In 1985 the axis was consumer versus office. In 2026 the mass market is a chat window with most of the connected world in it, and the professional market is people who need the thing to be right and will pay for that. The integrity animal takes the side where coherence is worth money. In the 1990s that was the design studio. Now it is the software team. The beak points toward the food, not toward the historical precedent.

The timeline is compressed. Steve Jobs was ousted in 1985 and returned in 1997: twelve years in exile. Sam Altman’s board crisis reached an agreement in principle after five days, and he formally returned twelve days after his removal. The whole drama runs roughly an order of magnitude faster, and this too is what you would expect. Machine lineages iterate in months rather than generations, and the institutions built around them inherit that clock. A fight that took Apple and Microsoft more than a decade to play out is being run at TechnoBiota speed.

The sizes are the wrong way around. In August 1997 Apple was close to bankruptcy, with a global personal-computer market share in the low single digits, when Gates appeared by satellite on a giant screen above Jobs at Macworld to announce a 150 million dollar lifeline, and the crowd booed. Jobs said the sentence he is still quoted for: “We have to let go of this notion that for Apple to win, Microsoft has to lose.” The integrity animal was nearly dead.

In 2026 the integrity animal is, by the latest reported annualized revenue figures, the larger of the two. If you were tracing, you would have gotten this wrong, and that is exactly the kind of error that tells you nobody is tracing. Convergent evolution predicts shapes. It says nothing about which occupant of a shape will thrive. The thylacine went extinct; the wolf did not. The niche guarantees that whoever survives in the predator slot will look something like a wolf. It does not guarantee the thylacine a pension. Apple in 1997 and Anthropic in 2026 have the same body plan and opposite fortunes, which is precisely what the mechanism says should be possible.

And Microsoft is in both stories. This one I did not see coming until I wrote it down. In 1997 Microsoft invested 150 million dollars to help keep the integrity animal alive, while a monopolist faced growing antitrust scrutiny and needed a visible rival. In 2019 Microsoft committed a billion dollars; it has invested more than thirteen billion in total since then, becoming the ubiquity animal’s patron. After OpenAI’s 2025 restructuring, Microsoft held roughly 27 percent of the company. Then it went further and became infrastructure for the other animal too: Microsoft distributes Claude through Azure and GitHub Copilot. The same organism from the earlier ecosystem has stepped into the new one and taken a different role. Not a competitor this time but a host. The old animals did not die. They became terrain.

Traits jump the fence

There is one more thing biology does that corporations also do, and it is my favorite because it is so shameless. In the biological world, traits mostly pass from parent to child. In the microbial world, and in the technological one, they also pass sideways: a bacterium picks up a useful gene from an unrelated neighbor and just keeps it. Horizontal transfer. TechnoBiota does this constantly, and the two animals are doing it to each other right now.

Anthropic released a small open standard in late 2024 called the Model Context Protocol, a way for a model to plug into tools and data. By spring 2025, OpenAI had added remote MCP support to its Responses API, and Microsoft had added support to Copilot Studio, VS Code’s agent mode, and Semantic Kernel. In December 2025 Anthropic donated the protocol to the Agentic AI Foundation, a Linux Foundation project co-founded by Anthropic, OpenAI, and Block. The integrity animal, of all things, produced the industry’s open plumbing, and the rivals now compete over the assistant while sharing its pipes.

Going the other way, in 2025 OpenAI agreed to acquire the hardware studio io in an all-equity deal valued at about 6.5 billion dollars. Jony Ive had founded it with former Apple colleagues, and his LoveFrom design firm stayed independent while taking design responsibility across OpenAI. Ive designed the iMac, the iPod, and the iPhone. Read that again as a biologist. The ubiquity animal has purchased the integrity animal’s most famous organ, from the earlier ecosystem, and is trying to grow it on its own body. Whether the graft takes is one of the more interesting open questions in the industry. Microsoft tried this repeatedly, with Zune and Surface and Windows Phone, and the organ mostly failed to take because the body around it was built for a different life. But traits do jump the fence, and sometimes they stick, and when they do the neat pair of shapes gets blurry for a generation before it re-sharpens.

The fence-jumping also exposes a real limit in the comparison. Apple pursued coherence by controlling hardware and software together. Anthropic distributes Claude through Amazon, Google, and Microsoft clouds, which is a considerable appetite for ubiquity, and pursues its coherence somewhere else: in the model’s behavior, which stays the same whichever machine it lives inside. Control of the experience and control of distribution turn out to be separable. A company can be particular about one and expansive about the other, and this animal is.

I want to be honest about what would falsify the picture. If OpenAI and Anthropic steadily converge on the same customers, the same commitments, and the same ways of selling, then I have exaggerated the pull toward two temperaments. Calling every awkward fact a borrowed trait would make the story impossible to lose, and therefore uninteresting to win.

Nobody decided this

Here is the part I actually care about, and it is where the essay stops being about two companies.

Every one of these moves has an author. Gates wrote the memo to Sculley; Apple refused to license; Amodei walked out; OpenAI acquired io. If you asked any of them, they would give you reasons, and the reasons would be good. But we have now watched the same two shapes emerge twice, from different people, in different decades, with different technologies, and the reasons were different each time. When the same form appears regardless of the reasons, the reasons are not the whole cause. The niche is doing some of the work. The people are how the niche gets what it wants.

I have argued elsewhere that a corporation is a TechnoBiota organism: it metabolizes, reproduces, competes, and persists beyond any of the humans inside it, and it selects for the human behaviors that serve its growth. This is what that looks like from the outside. The founders experience themselves as choosing a strategy. From the ecosystem’s point of view a strategy is a phenotype, the market runs selection on phenotypes, and the phenotypes that keep surviving in a general-purpose-technology niche wide enough for two are the ubiquity animal and the integrity animal. Jobs was a specific and irreplaceable person. So was the thylacine. The wolf-shape did not need either of them.

A mechanism, unlike an analogy, tells you what to watch. Watch the costly decisions. Does a company protect a promise when its biggest customer asks it to bend? Does the refusal attract enough loyalty to pay for itself? Does a push for reach change who gets hired and promoted? Those moments reveal the animal’s emerging shape better than the adjectives on its website. Displacement says the two will keep pulling apart along whatever axis remains: expect the ubiquity animal to get more promiscuous, more ad-funded, more embedded in everyone’s everything, and the integrity animal to get more expensive, more selective, and more insufferable to those who are not in the cult. Convergence says the third entrant into the niche, Google, which is trying to be both animals at once, faces Gause’s problem in its purest form and will have to pick a shape or partition itself. And the fossil record says the pair is stable only until the niche itself moves. The personal computer duopoly dissolved not because either animal beat the other but because the niche walked to the phone, and only one of the two, the one that had spent twenty years practicing coherence, managed to carry its body plan from the desk into the pocket.

The names on the buildings are new. The founders are new. The technology is genuinely, alarmingly new. But the two shapes are old, older than either company, older than computers, and they will be here, wearing new names, the next time a general-purpose technology opens a niche wide enough for two.

The next time a founder takes the stage and tells you what their company believes, listen. They may mean every word. Then look at the customers, the contracts, the hiring, and the competitor across the street, and ask what rewards that belief, what makes it expensive to abandon, and what would happen to someone who tried.

We keep thinking we are watching people make history. We are watching a habitat make animals.