The Autonomous Enterprise Is Weiser's Third Wave

In September 1991, Mark Weiser published an essay in Scientific American that described the shape of mature computing more precisely than anything that would follow it for a decade. He opened with a sentence that has been quoted in every technology publication of consequence since:

"The most profound technologies are those that disappear. They weave themselves into the fabric of everyday life until they are indistinguishable from it."

— Mark Weiser, The Computer for the 21st Century, 1991

The name Weiser gave to the pattern in 1991 was ubiquitous computing. He also called it embodied virtuality — the state where computer-readable data has been brought into the physical world and stopped being noticeable as data. Four years later, writing with John Seely Brown, he would give the aesthetic a second name: calm technology. The two words got attached to the same idea, and both stuck.

What Weiser meant was not what most technologists heard. He was not describing the next generation of computers. He was describing a change in the relationship between computers and the humans who use them — a change so complete that computers, as a category, would stop being the thing anyone paid attention to.

Weiser''s three waves

Weiser argued that computing had come in waves, and each wave was a different pattern of attention.

The first wave was the mainframe. One large machine per organization. Many users, all coming to it. Attention flowed from human to machine.

The second wave was the personal computer. One machine per person, on your desk and then in your bag. Attention still flowed to the machine. The distance had collapsed; the pattern had not.

The third wave was ubiquitous computing. Weiser was specific about the scale: not one machine per person, but hundreds of machines per room. He estimated a typical room of the future might hold around 100 inch-scale devices ("tabs"), 10 to 20 foot-scale devices ("pads"), and one or two yard-scale displays ("boards"). Attention would not flow to any of them. It would flow through them, and past them, to whatever a human was actually trying to do.

He was careful to insist that the third wave would not look like the second wave scaled up. A better personal computer would still be a personal computer. Ubiquitous computing was a different pattern.

Weiser''s genealogy

The 1991 essay was more philosophically grounded than most people who cite it seem to remember. In the space of one paragraph, Weiser lists five thinkers he considers his lineage:

• Herbert Simon on "compiling" — the way learned tasks retreat from conscious effort.
• Michael Polanyi on "the tacit dimension" — the knowledge we act on without being able to articulate.
• J. J. Gibson on "visual invariants" — the perceptual constants we use without noticing them.
• Martin Heidegger and Hans-Georg Gadamer on the ready-to-hand and the horizon — tools we operate through rather than on, and the frame of reference that recedes when we focus.
• John Seely Brown, then at PARC, on the periphery — attention that is present without being central.

His summary of what all five have in common is the clearest one-sentence statement of what mature technology feels like: things disappear in use when we are freed to use them without thinking, so that we can focus beyond them on new goals.

This is the intellectual lineage Enterprise Autonomy inherits. Not as a marketing move — as a genuine claim about the shape of mature enterprise computing.

The two analogies that survive translation

Weiser used two analogies to explain how a technology becomes background. Both are load-bearing in the essay, and both translate almost perfectly to what is happening in enterprise software right now.

The electric motor. In 1900, a factory ran on a single large engine that drove workstations through a system of shafts and pulleys. The engine was the technology. Everyone in the factory knew where it was. By the time Weiser was writing, the shop manual of a typical automobile listed 22 motors and 25 solenoids — one under each seat, one in each window, one at the wipers, one at the fuel pump, one at every point in the vehicle where motion was needed. The driver could, with careful attention, notice any one of them. There would be no point in it. Motors had stopped being a technology and had become a component of everything the car did.

Enterprise AI is on the same trajectory. In 2023, a Fortune 500 industrial company had "an AI initiative" — one program, one budget, one steering committee. In 2027 the same company has forty autonomous workflows. Nobody attends the steering committee anymore, because there is no steering committee. AI has become a component of every operation the business runs. The word has begun to disappear from the internal vocabulary. The workflows are known by what they do — nonconformance triage, cash application, freight rebooking — not by what they are.

Literacy. Weiser calls writing "the first information technology." He points out that in industrialized countries, writing is so ubiquitous — in books, on street signs, on billboards, on shop signs, on candy wrappers, on graffiti — that we no longer notice it. The information is ready for use at a glance. Reading has become invisible to the reader.

Enterprise Autonomy is the operational version of literacy. When a business function operates through autonomous workflows, the workflow stops being visible. The plant manager sees yield. The controller sees the close. The supply-chain planner sees inventory position. What happens in between — the reading, the deciding, the doing — has been absorbed. The outcome arrives. The mechanics recede.

At-hand, not in-focus

The most common objection to this framing goes: "But our AI is very visible. It talks to us. It asks follow-up questions. It clearly is not disappearing. So the theory must be wrong."

The theory is fine. The AI is wrong. Or, more precisely, the deployment pattern is wrong.

Weiser, and Heidegger before him, made a distinction between two ways a tool can be present. In-focus is when the tool demands your attention: you interact with it as a thing. At-hand is when the tool has receded to the periphery: you use it through rather than at. You know your phone is working when you pick it up; you are not staring at it. You know your car''s anti-lock brakes are working when they work; you are not managing them.

Copilots and chatbots, as they are commonly deployed in 2026, are the opposite of at-hand. They are perpetually in-focus. They demand engagement to produce value. When engagement stops, output stops. That is not a defect of the underlying language models. It is a defect of the pattern in which they are being deployed.

An autonomous workflow that runs quietly, catches an exception, escalates when appropriate, and lets a human review the escalation at their leisure — that is at-hand not in-focus. That is the third-wave pattern. A chatbot that a human has to interrogate to get answers is not. The test is simple: if a human stopped opening the interface for a month, would the workflow keep producing value? If yes, third wave. If no, second wave with third-wave marketing.

Weiser''s warning

Ubiquitous computing is not neutral. Weiser was explicit about this. He wrote that hundreds of computers in every room, all capable of sensing people near them and linked by networks, have the potential to make totalitarianism up to now seem like anarchy. He also wrote that cryptographic techniques already exist to prevent the outcome, but only if they are designed in from the start.

The same warning applies, one wave later, to Enterprise Autonomy. Sixty autonomous workflows making decisions that touch customers, suppliers, and financials without a policy fabric, an evidence layer, and independent audit is not the third wave. It is a governance disaster with better latency. Weiser saw this in 1991 for a world of ambient computers. It is the same problem, one wave later, for a world of ambient agents.

Any vendor or leader who describes Enterprise Autonomy as "no humans" is either misunderstanding it or selling something. The correct posture is: fewer humans in the mechanics, more humans on the outcomes, more governance around the autonomy, not less.

What follows for the market

If Enterprise Autonomy is Weiser''s third wave for the enterprise, three things follow.

It is not a product category. Third waves are not categories. They are what mature computing looks like. Vendors that treat it as a category will lose to vendors that treat it as an operating pattern.

It is architectural, not tactical. You cannot procure your way into the third wave. It arrives through a series of architectural commitments — where state lives, how agents authenticate, how the reasoning core connects to the systems of record, how autonomy is audited — and once those commitments are made, they compound.

It is measured in outcomes, not in productivity. The change in the unit of account — from minutes-saved-per-employee to yield, margin, throughput — is the tell that the wave has changed.

Weiser''s fate, and ours

Weiser wrote that ubiquitous computing would gradually emerge as the dominant mode of computer access over the next twenty years. He also wrote that a business world with computers embedded in it, rather than sitting on it, would come to feel more human than the world before it. His final image is the one worth carrying: machines that fit the human environment instead of forcing humans to enter theirs will make using a computer as refreshing as taking a walk in the woods.

The consumer version of this arrived, roughly on his timeline, between 2005 and 2015. It is now the ordinary condition of consumer computing. Most people cannot describe what changed. That is what a successful third wave looks like: it becomes invisible in use.

The enterprise version is arriving now. The organizations that recognize the shape and act on it will look, three years from now, less busy and more effective than their competitors — for reasons their competitors will find difficult to articulate. Their operating leverage will expand quietly. Their AI, if you can still find it in their internal vocabulary, will have stopped being called AI.

Weiser''s essay looked visionary in 1991 and quaint by 2015. This one, if we are lucky and honest, is on the same trajectory. Obvious in retrospect. Not yet visible in the moment. Not yet — but any day now.