Why the Future Always Looks Like the Past

At Saqqara, on the desert plateau west of Cairo, there is a door that has been standing open for nearly five thousand years. It cannot close. It was never supposed to.

The door belongs to the funerary complex of the pharaoh Djoser, built around 2670 BC and generally attributed to Imhotep. The complex is one of the earliest experiments in monumental architecture made from dressed stone, a technology so consequential that once humans learned to build this way, architecture acquired something close to immortality.

Stone was the breakthrough, and almost immediately the builders used it to pretend they were not building in stone. Columns were carved to resemble bundles of tied reeds. Limestone ceilings were shaped like wooden beams. Blue-green tiles reproduced the reed matting of an earlier palace. And near the entrance, masons cut enormous false doors into rock with almost absurd fidelity to the wooden originals. One of them stands ajar. They even carved the hinge.

A hinge. In limestone. A mechanism whose entire purpose is movement, reproduced in a material that guaranteed it would never move at all.

It is tempting to look at this and see timidity: the builders had acquired a radical new material but apparently lacked the imagination to use it on its own terms, and were still thinking in reeds and timber. That explanation becomes less satisfying the longer you look at the history of technology, because we keep doing this.

The first printed books looked handwritten. Much early cinema looked like theatre. Television borrowed radio programmes. Computers became offices filled with files, folders, desktops and wastebaskets. Smartphone buttons once gleamed like plastic and leather. Electric cars are legally required, in some circumstances, to make noises reminiscent of combustion engines.

And now artificial intelligence—a technology routinely described as transformational—is arriving as a chat window, a search box, a copilot, an assistant, an employee and, increasingly, a little simulated person operating a computer with a simulated mouse.

There is a word for this family resemblance between a new thing and an older one: skeuomorph.

And if the story stopped there, this would be a familiar essay. New technologies begin by imitating the technologies they replace. Eventually we learn what the new medium is really for, the imitation falls away, and the technology becomes native. The car loses the horse. The computer loses the paper. AI loses the chatbot.

It is a wonderful story. It is also, in several important ways, wrong.

The word skeuomorph was coined in the late nineteenth century by Henry Colley March. Its roots are Greek: skeuos, an implement or vessel, and morphē, form.

March was interested in ornament that had once been structure, a feature necessary in one material and copied into another after its original function had disappeared. Imagine a wooden beam translated into stone: the beam's shape survives, its job does not.

That is narrower—and more interesting—than the way the word tends to be used in technology now, where almost anything vaguely old-fashioned can be labelled skeuomorphic. Fake leather is skeuomorphic. A floppy disk save icon is skeuomorphic. A podcast app resembling a reel-to-reel machine is skeuomorphic. March was after something more specific: the ghost of an old constraint preserved inside a new material.

There is another wrinkle. March coined the word inside a Victorian intellectual world obsessed with evolution, where scholars routinely described cultural objects as passing through stages resembling origin, development and decay. The skeuomorph was something to classify, explain and, in one later scholar's description of March's project, to "tame." Which means the familiar idea that technology begins by imitating and then matures out of imitation was not attached to skeuomorphism later. It was there near the beginning.

That evolutionary story still shapes how we talk about technology. We laugh at the fake leather once the leather disappears, we call early websites "brochureware," and we describe AI products as horseless carriages. Implicit in all of these jokes is a belief that imitation is a kind of embarrassment: an understandable but primitive stage before designers finally discover what the technology wants to become.

Archaeology gives us reason to be much less smug.

In 2015, anthropologist John Blitz argued that skeuomorphism could actually be a cause of technological adoption, and his evidence came from pottery. Across cultures, early ceramic containers frequently retained forms and textures associated with the baskets, gourds and other vessels that preceded them. Blitz's point was that these similarities need not represent a failure to imagine a properly ceramic pot. The resemblance itself could make the strange object comprehensible.

An unfamiliar technology gets to borrow the reputation of the familiar one. It enters an existing category of value, evokes memories and rituals already attached to its predecessor, and gives people some idea of what the object is, what it is worth and how they are supposed to behave around it. The skeuomorph is not necessarily baggage. It can be the sales pitch.

Blitz is careful about the limits of this. It explains adoption, not invention. Basket marks do not explain why somebody discovered pottery; they help explain why somebody else wanted the pot.

Archaeologist Catherine Frieman pushes the argument further. She is suspicious even of the assumption that imitation needs to be justified as a functional bridge to innovation. Imitation, in her account, can carry identity, belonging, status and cultural meaning of its own, and need not be a temporary failure to become something else.

More unsettlingly, Frieman has shown how easy it can be to get the direction of imitation wrong. Archaeologists once interpreted certain prehistoric stone artefacts as copies of prestigious metal originals; her work suggested that in many instances the early metal objects had themselves been shaped in forms already socially meaningful in stone. Experts had the skeuomorph backwards.

That should impose some humility on anyone declaring, in 2026, that the chatbot is obviously a skeuomorph of conversation or the AI agent obviously a skeuomorph of an office worker. We have been wrong about objects we can dig out of the ground. We may be even worse at diagnosing a technological transition while standing inside it.

Around 1455, Johannes Gutenberg finished one of the most consequential manufacturing projects in human history. And then he sent the product away to be finished by hand.

Early printed books are almost comically deferential to manuscripts. The typeface of the Gutenberg Bible reproduced the dense blackletter forms readers already associated with scribes. The pages contained spaces deliberately left empty so rubricators could add initials and headings in coloured ink. Illuminators decorated borders, and binders customised copies for their owners. The printing press—the machine that would help industrialise information—produced sheets designed to be completed by the technology it was supposedly replacing.

Again, the easy interpretation is that Gutenberg had invented the future but lacked the nerve to make it look futuristic. A better interpretation is that nobody was shopping for "the future." They were shopping for books. The first job of the new technology was not to announce its rupture with the old world; it was to make the rupture survivable.

Print did not invent every navigation device that later became associated with it, but mechanical reproduction standardized and amplified them. The title page became a regular feature. Printed pagination spread. Tables of contents and indexes became more powerful because multiple copies could share stable locations, so a reference could point you to the same place in your book and mine.

The deeper innovation was not that printed books eventually stopped looking handwritten. It was that the medium acquired a new grammar of navigation. The visible skeuomorph was typography; the more important change was what a book could now do.

This distinction turns up repeatedly. Much early cinema parked the camera in front of performers as though it were a particularly good theatre seat: frontal, static, actors entering and leaving a frame that behaved like a stage. Then, piece by piece, cinema developed a syntax no theatre could reproduce. Cut here. Move closer. Show a face, cut to the thing they see, return to the face, and put two events happening miles apart beside each other in time.

One lovely artifact of the transition appears in The Great Train Robbery from 1903: the famous close-up of an outlaw firing his pistol directly toward the camera. The shot was so detached from the narrative that exhibitors could place it at the beginning or the end of the film. The image did not yet know where it belonged. Cinema had invented the shot before it had completely invented the sentence.

Then there is recorded sound. For decades after Edison recorded "Mary Had a Little Lamb" in 1877, much of what people imagined doing with recording technology consisted of capturing things that already existed: speeches, lectures, poems and books. Newspapers imagined literature sold by the "bottle," and Edison claimed entire novels would fit on a few phonograph cylinders, considerably before the technology could actually manage it.

You could tell this story as yet another medium failing to become itself, except that the talking book never went away. Nor should it have. The first novel-length talking books were created for blind readers, including people who had lost their sight in the First World War. For them, reproducing the old form was not an embarrassing transitional use of the new medium. It was access.

This is an important correction to the whole skeuomorphic story. Sometimes the borrowed form stays because users have not yet learned the new one. Sometimes it stays because it works. Sometimes it stays because it means something. And sometimes removing it closes a door.

The most useful story about skeuomorphism does not involve an interface. It involves a factory ceiling.

Before electricity, industrial power had a geometry. A factory might have one large steam engine turning a long overhead driveshaft, with belts descending from that spinning shaft to individual machines. The arrangement dictated the building: machines had to be placed according to their access to power, and long shaft runs lost efficiency, so factories tended to be built vertically with machinery clustered around the transmission system. The architecture followed the engine.

Then electricity arrived, and surely this is the point at which the factory becomes electric.

Not quite. Factories replaced the old central prime mover with electric motors while preserving much of the existing shaft-and-belt system. There was even a name for the compromise: group drive. Electric motors powered sections of shafting which continued to distribute energy mechanically to groups of machines. The source of power had changed. The topology had not.

It is difficult to imagine a cleaner physical skeuomorph. Electricity had been installed inside a factory still organised around steam, and for years productivity barely responded.

Economic historian Paul David would later make this one of the classic examples of why general-purpose technologies can take astonishingly long to show up in productivity statistics, and his timeframe is worth getting right. It was roughly four decades from the opening of the first central power station in 1882 to the point in the early 1920s when electrification had achieved enough diffusion and reorganisation to begin showing its larger effects.

The delay was not simply that factory owners were foolish. Keeping existing equipment could be economically rational: machinery represented sunk investment, and retrofitting incrementally was cheaper than rebuilding everything at once. Engineers could already imagine many of the benefits of electric power. What was missing was not a brilliant PowerPoint presentation about the factory of the future. It was years of co-invention.

The breakthrough was unit drive. Instead of using motors to turn inherited shafting, manufacturers began giving individual machines their own electric motors. That sounds like a small engineering change. It rearranged the factory. Machines no longer needed to orbit a power-transmission system, so buildings could spread horizontally, production could be laid out according to the movement of materials, departments could stop independently, and equipment could be rearranged more easily.

And then comes my favourite detail in David's paper. Factories could have skylights. The old overhead belts and shafts had occupied the ceiling and filled workshops with dust; remove them and suddenly the roof could admit daylight.

Think about the causal distance here. Electricity leads to individual motors. Individual motors eliminate a driveshaft. No driveshaft means different buildings, which means different layouts, which means different production systems. And somewhere four steps away from the original invention, someone gets a window.

Nobody looking at an electric motor in 1890 would have put "sunlight" near the top of the feature list. Yet sunlight was part of what the technology eventually made possible.

This is the moment where skeuomorphism becomes more useful than an aesthetic joke, because the problem with the old factory was not that the driveshaft looked old. The problem was that everything else had been forced to organise itself around the driveshaft. The shaft coupled decisions together: where machines stood, how buildings were shaped, where materials moved, how maintenance happened, what the roof could contain. Replace the engine while keeping that coupling and surprisingly little changes. Remove the coupling and the whole system can move.

So here is a more useful question than Does this technology still resemble the thing it replaced?

What else is forced to remain true because this thing is still here?

If the answer is "almost nothing," you may be looking at harmless inheritance. If the answer is a chain of dependencies, you may have found the expensive kind.

For the rest of this essay it is useful to give these inheritances names. They are not established scholarly categories, just a lens for this argument.

Call the first ornament: visible inheritance with few downstream dependencies. Fake leather. The floppy-disk icon. The carbon copy hiding inside the letters CC.

Call the second structure: inherited arrangements that other decisions are forced to obey. The driveshaft. The file hierarchy. Perhaps the org chart.

And add a third, doctrine: an inherited belief that made sense under an old constraint and keeps directing behaviour after the constraint has disappeared.

The visible skeuomorph attracts design essays. The structural one eats decades. The doctrinal one can survive indefinitely, because eventually everybody forgets there was ever another reason for doing things that way.

In 1981, Xerox released the Star. Its designers faced a formidable problem: how do you persuade an office worker to use a computer when almost none of the vocabulary of graphical computing exists yet?

Their answer was audaciously conservative. Put the office inside the machine. A desktop, documents, folders, file drawers, an inbox and a wastebasket.

The metaphor was deliberate. Xerox was selling an automated office to people who already understood physical offices, which is skeuomorphism doing precisely the job Blitz described: making an unfamiliar system legible by placing it inside a familiar world.

It worked almost too well. In 1998, Star designer David Canfield Smith looked back at the desktop metaphor and complained that computing had not escaped it. The metaphor, he said, had been "too successful for too long" and accumulated "too much inertia." He wanted computers organised more around information retrieval. Search, in other words.

That comment now reads like a message sent forward in time. A generation later, professors began noticing that some students did not understand what they meant when they asked, "Where did you save the file?" It was not that the students had forgotten the location. The question itself was becoming alien. People raised on global search, cloud libraries, app silos and recents views did not necessarily construct the same mental tree of nested directories earlier computer users had learned.

Notice what happened there. The folder icon survived and the folder idea weakened. The ornament outlived the structure.

This is why the great rebellion against skeuomorphism in interface design now looks slightly misplaced. When Apple stripped away much of its mimetic surface detail in 2013—leather textures, bevelled controls and ornamental realism—the change was widely read as a victory for flat design. But Apple's own iOS 7 guidance still emphasized translucency, visual layers, realistic motion and depth. A folder with no cardboard texture is still a folder, and a "page" rendered in perfect geometric minimalism is still a page. The realistic surface receded while the inherited conceptual architecture remained largely untouched. We won the war against fake stitching, and the office was still inside the machine.

Sometimes even the ornament turned out to have been doing something useful. Flatten interfaces aggressively enough and buttons begin to resemble labels; visual signifiers disappear. Twelve years after iOS 7, Apple's Liquid Glass made material cues, layering and depth overt again—not as a revival of stitched leather, and not as the first return of depth, but as evidence that pure abstraction was never an obvious final state.

The arc, again, is not skeuomorphic → native. It may be closer to borrow → abstract → discover what the borrowing was accomplishing → put some of it back.

At this point the word native starts becoming suspicious. Native to what? A screen? A processor? A human hand? A business model? A medium?

Media theorists Jay David Bolter and Richard Grusin offer a particularly inconvenient answer. Their theory of remediation argues that media acquire meaning by continually refashioning other media. Photography remediates painting. Cinema remediates photography and theatre. Television remediates cinema and radio. Digital media remediate almost everything. And importantly, the process also runs backwards, with old media changing in response to new ones. There is no clean finish line where a medium becomes historically pure.

History supplies some wonderfully awkward examples. Consider QWERTY, whose layout comes from the mechanical era of the typewriter. Whatever argument one accepts about why the arrangement prevailed—and the famous lock-in explanation is itself contested—the form survives long after the relevant mechanical machinery disappeared. We do not wake up each morning furious that our keyboard has failed to achieve its native digital form. It is simply a keyboard. The metaphor has been naturalised.

Then consider VisiCalc, the spreadsheet that helped transform the personal computer from hobbyist equipment into a business machine. Its rows and columns came directly from the paper accounting ledger, and its name meant "Visible Calculator." Two inherited metaphors. But its defining capability—change one cell and automatically recalculate everything dependent on it—had no paper equivalent. VisiCalc was skeuomorphic and unprecedented on the same day. There was no childhood followed by adulthood. The inherited form was precisely what made the alien capability comprehensible.

And sometimes genuinely new forms do appear. In 2008, Loren Brichter needed a way to refresh the feed in his iPhone Twitter client Tweetie. Instead of putting another button on the screen, he made the content itself become the control: pull the list beyond its upper boundary, release it, and the feed refreshes. Pull-to-refresh. No lever, no dial, no desktop object, no paper ancestor. It is difficult to think of anything the gesture is pretending to be.

Which matters, because it reveals something the phrase "technology eventually finds its native form" conceals. Technologies do not discover forms. People invent them. Someone has to have the idea.

There may therefore be no Platonic interface waiting for AI somewhere in the future. There will be things people build. Some will stick and some will fail. Some will imitate the past, and some will look unprecedented for ten years and then become so obvious that our children will assume they always existed. "Native" may simply be what a metaphor feels like after everybody forgets it was a metaphor.

Look at contemporary artificial intelligence and the skeuomorphs are everywhere. It is a chatbot, borrowing the rhythm and visual grammar of messaging. It is a copilot, sitting metaphorically beside you. It is an assistant with a name, one that types and tells you it is "thinking." Companies describe software agents as employees, give them roles and objectives and occasionally put them on organisational charts.

And one of the strangest new classes of AI system is the computer-use agent: a machine that takes screenshots of ordinary software, visually identifies buttons and fields, then moves a pointer around the interface as though it were a person sitting at the computer.

Pause on how bizarre that is. In 1981, Xerox put an office inside the computer so human beings could understand the machine. Forty-five years later, AI systems are learning to impersonate human beings so they can operate the simulated office. The computer put on a costume so we could use it. Now we are the costume so it can use the computer.

The loop has closed. A folder icon imitates cardboard for the benefit of a human who may no longer be looking at the screen, and the AI imitates the human for the benefit of software that never needed cardboard in the first place. Researchers working on agent interfaces have already begun describing this mismatch explicitly: graphical software is a dense, culturally learned presentation layer built for human eyes and hands, not necessarily a clean representation of the actions a machine can perform.

Below that human layer, another architecture is appearing: protocols by which AI systems can interact with tools and data more directly. No eyes, no mouse, no little person clicking menus.

It is very tempting to declare the GUI-driving agent AI's horseless carriage, and maybe it is. But history tells us to ask a more precise question. Not does this resemble an old thing? Ask: what is coupled to it?

A computer-use agent is tightly coupled. Its abilities depend on what is visible on screen, which depends on where a designer placed a control, which in turn depends on the application. Change the layout and the automation can fail. The inheritance is not simply aesthetic: the machine's action space has been forced through an interface designed for fingers. That looks much more like a driveshaft.

But even here, the old system has a defence. A human interface is inspectable. You can watch an agent work, you can stop it, and you can see the field it is changing. A direct machine-to-machine action might be more efficient while being much harder for the accountable human to witness. Sometimes the inefficient skeuomorph may remain because visibility matters. The old shape can be a control surface.

This is the question that matters. Most debates about "AI-native" software concentrate on the obvious costume: chat, prompt boxes, typing animations, anthropomorphic names. These may all be worth questioning. But the factory story suggests we should spend less time staring at the horse head and more time looking for the shaft running through the ceiling.

What has AI inherited that forces everything around it to stay arranged in an older pattern? There are several candidates.

The document. A PDF is an extraordinary machine for pretending a sheet of paper still exists. It preserves dimensions, visual coordinates, columns, headers and footnotes because one of its great historical purposes was ensuring a document could be viewed and printed consistently across different machines. Models now expend real computation reconstructing the logical relationships hidden inside those visual coordinates. The page was once an interface designed for eyes; now machines reverse-engineer the page to recover the information underneath it.

The application. Software remains divided into separate walled products, each with accounts, permissions, controls and interfaces. An AI acting across them often has to traverse boundaries originally designed around a person opening several programs.

The working day. Our coordination systems are organised around an animal that sleeps: meetings, business hours, response latency, the Monday morning inbox. These may be perfectly sensible when the bottleneck is the availability of humans. They become stranger when part of the work can continue indefinitely.

And then there is the most provocative candidate: the job.

Companies are beginning to describe AI systems in the language of employment, with roles, responsibilities, onboarding, performance, colleagues and headcount. A 2026 working paper surveying 1,261 managers found that 23 percent worked in organisations where AI agents had already been formally represented on organisational charts.

The same research contained a more subtle warning about what this metaphor can do. Across the experiment as a whole, effects on error catching were small. But among managers at organisations that had already institutionalised "AI employees," describing identical output as coming from an AI employee rather than an AI tool reduced monitoring intensity by 16 percent and shifted some perceived accountability away from the manager. That is a subgroup result in a working paper, not a universal law, and it has sometimes been overstated in secondary coverage. But the qualification makes the finding more interesting, because the effect appeared where the metaphor had already become institutional.

Run the driveshaft test. If work is organised into job-shaped boxes, what else gets coupled to those boxes? Headcount, budgets, salary bands, procurement, permissions, accountability, working hours, management layers, performance reviews, seat-based software licences and legal responsibility.

A job is not merely a word. It is an organising primitive around which enormous portions of the modern corporation have been constructed, which means an "AI employee" may be skeuomorphic in a much deeper sense than an assistant icon. The org chart may be the line shaft. Not because it is old, but because so many unrelated decisions have been chained to it.

There is another inheritance hiding here. Not a visual metaphor, and not an organisational component, but a belief.

Consider the phrase above the fold. It came from newspapers displayed folded on newsstands, where the upper half of the front page received special attention because it was physically visible to passers-by. Then the language migrated to web design. There is no physical fold on a screen, but the doctrine survived.

Eventually the reaction against it became simplistic too. Data often cited to prove that users happily engage "below the fold" actually showed something more complicated: attention peaked slightly beneath it and then declined. One inherited rule got replaced by another inherited rule.

That is the dangerous thing about doctrine. You can remove the original constraint without removing the conclusion people learned from it, and AI is likely to produce dozens of these.

One possible example is the belief that oversight is a step. Human organisations often place quality control at a handoff: write, submit, review, approve, done. That makes sense in a process defined by work moving between people. But an automated process need not treat oversight as a gate at the end. Checking can be continuous, sampled, adversarial, redundant or embedded into the production process itself. Manufacturing spent much of the twentieth century learning to build quality into production rather than depend on inspection at the end, and software engineering learned an analogous lesson through automated testing and continuous integration. Yet much of the conversation around supervising AI still imagines a person standing at the end of the conveyor belt, inspecting completed output.

Perhaps that is necessary. Perhaps it is inherited doctrine. The important thing is to notice that these are different questions from whether the chatbot should have speech bubbles.

There is, finally, a serious argument that turns this entire essay around.

In 1979, media theorist Paul Levinson proposed the idea of anthropotropic media evolution. Media, in his theory, tend to move toward the human. Telegraph gives way to telephone because people prefer speech to decoding dots. Black-and-white imagery gives way to colour because human vision perceives colour. Technology succeeds partly by extending human sensory and communicative behaviour rather than escaping it.

Take that theory seriously and AI's human resemblance begins to look very different. Maybe conversation is not a training wheel. Maybe natural language is not a primitive interface waiting for something more "AI-native." Maybe AI is the most anthropotropic medium ever invented: a machine whose great breakthrough was precisely that humans no longer had to translate themselves into machine language before asking a machine to do something. On this reading, predicting the death of chat might be like predicting that telephones will eventually mature beyond carrying voices.

But Levinson's argument also suggests a useful distinction. Speech is human; an org chart is not. Natural language is ancient; a per-seat software licence is not. Conversation reflects something deep about human communication, while a 9-to-5 working day reflects the rotational period of the Earth, labour law, industrial history and the fact that mammals become tired.

We should therefore be careful about putting every human-shaped feature of AI in the same bucket. The sensory inheritance may be durable and the institutional inheritance contingent. Natural language may survive where the employee metaphor does not. The conversation may stay while the job disappears around it.

A reader in Mainz in 1455 could understand the printing press in an afternoon. It made books faster. What that reader could not see was everything that had not been invented or standardized yet: the modern title page, reliably paginated editions, stable cross-copy references, mass-circulation periodicals, or the publishing and reading practices that would grow around them. The reader was not stupid for failing to predict these things. They were absent. There was no angle from which to see them.

A factory owner in 1900 could see electric motors. What they could not simply buy off a shelf was the complete reorganisation of architecture, management and production that would eventually make those motors transformative. The lag was not merely a shortage of imagination. It was the time required to invent everything around the invention.

That may be the most important historical lesson for AI. We can see what today's systems do: converse, write, generate images, manipulate software, search, reason and automate fragments of work. What we cannot yet see is which of today's forms are equivalent to Gutenberg's handwritten typeface—temporary familiarity waiting to fall away—and which are equivalent to the page itself, inherited structures so useful that nobody will bother replacing them.

We cannot know which conventions are scaffolding and which are permanent. And we should be particularly suspicious of anyone who claims to know what the "native" form of intelligence on a computer inevitably looks like, because there is no evidence that technologies contain a finished design inside them, waiting to be discovered.

Forms are invented. Constraints disappear. Metaphors survive. Old media mutate and new media imitate them. People defend apparently obsolete conventions for reasons that turn out to be excellent. And occasionally an inherited driveshaft sits above everybody's head for forty years because removing it requires rebuilding the room.

Which brings us back to the name of this age. "The Skeuomorphic Age" sounds as though it describes a temporary period, a larval stage: soon AI will cease imitating people, applications, offices and documents, it will become itself, and the future will finally look like the future.

But perhaps there is no such moment. Perhaps there has never been a non-skeuomorphic age. The desktop was once a metaphor and is now furniture. The page was once a physical constraint, and now people talk about a "web page" without picturing paper. Children tap a floppy disk to save work they have never stored on a floppy disk. We CC colleagues without carbon. We open windows that contain no glass. The metaphors stop feeling metaphorical because the original joke disappears from living memory.

Every skeuomorph is a theory about what a new technology is for. The reed column was a theory about stone. The manuscript typeface was a theory about print. The desktop was a theory about personal computing. The AI assistant is a theory about intelligence, and the org chart is another. Most of these theories are not completely wrong. They are simply smaller than the technologies eventually become.

Somewhere at Saqqara, a stone door remains permanently open. Its builders had just been given one of the most transformative materials in architectural history, and they used it to reproduce, with obsessive care, the hinge of a wooden door. They were not fools. They were standing at the beginning of something enormous, trying to understand the future using the shapes the past had given them.

So are we.