In the 1990s, Bill Gates described the “information highway” in The Road Ahead. The image was not only PC and Windows marketing. It was a bet on civilizational architecture: the network would become the road; the computer would leave the center of the desk; and, in the limit, it would stop existing as a social category. You would not “use a computer” in the same way that, after electrification, almost nobody “uses a generator”: they simply inhabit a world that electricity makes possible.
Three decades later, with frontier models, agents, and natural-language interfaces, that disappearance no longer sounds like CD-ROM futurism. It has a concrete outline. Not because silicon vanished. Because the conscious act of computing is leaving the center of experience.
What the prophecy actually meant
There is a useful confusion to clear up. “Computers will cease to exist” almost never meant shutting chip factories. It meant, in stacked layers:
- Object disappearance: the PC stops being the only digital altar.
- Place disappearance: the network and services, not the local machine, become the axis.
- Attention disappearance: the interface recedes; the service embeds in cars, homes, pockets, offices, and routines.
Mark Weiser called this ubiquitous computing: the deepest technologies are the ones that disappear. Gates translated the intuition into market language and the “information highway.” The road mattered more than the car parked in the desktop garage.
What materialized between 2000 and 2020 was the easier half of the thesis: smartphones, cloud, apps, weak assistants, IoT. The computer multiplied and shrank. Even so, experience remained mediated by explicit software: icon, menu, form, click flow. The object changed shape; the ritual of operating the system remained.
The leap AI introduces
Frontier AI shifts the location of the interface. It is not enough for the device to be small or ubiquitous. What changes is how humans formulate work.
| Phase | Where the center sits | What the human does |
|---|---|---|
| Desktop | Machine and files | Operates the software |
| Cloud and app | Service and account | Navigates flows and screens |
| Model and agent | Intent and context | States the goal; the system orchestrates |
Before, you modeled the problem inside the software. Increasingly, you describe intent in language. The product stops being the app and tends to become the agent (with memory, tools, permissions, and an action trail). The interface stops being only screens and menus and becomes conversation, context, and execution.
It is the same movement Gates described, one level up:
- 1990s: make the terminal stop being the center of digital life;
- 2020s: make the explicit workflow stop being the center of cognitive work.
The “information highway” was the network. AI is the vehicle that, on many stretches, plans the intermediate route. In some cases, it already accelerates on its own.
Three layers that finally aligned
The 2025–2026 glimpse does not come from one model. It comes from three pressures lining up:
-
Model as cognitive operating system
The place where work happens stops being the isolated application and becomes the session with the model: long context, tools, agents, memory. The “system” that matters is the one that understands intent and decomposes it. -
Action, not only text
Generating a paragraph is the floor. Value moves to code, calendars, browsing, APIs, queues, and then to robotics and physical end effectors. AI leaves the chat box and tries to operate the world. That theme also runs through Elon Musk’s July 2026 interview with The Economist: digital intelligence without a body is only half the problem; physical intelligence completes the economics of abundance. -
Falling marginal cost of intelligence
When reasoning, drafting, summarizing, and orchestrating become cheap enough, menus and forms become friction. The market pushes the disappearance of “computer-as-work” for most intermediate tasks.
Gates spoke of access and network. AI speaks of delegation. The road existed; what was missing was a driver that understood the destination in natural language.
What disappears and what multiplies
Disappearance is social and experiential, not material. The honest map is almost the opposite of the slogan:
- More chips, more data centers, more electricity and cooling.
- More identity, permission, token budgets, logs, and governance.
- More invisible software under every natural sentence.
The “computer that disappears” is, in practice, the total computer: ubiquitous, metered, billed, and geopolitical. The interface recedes for the user; the infrastructure shouts at the network operator, the regulator, and the CFO.
There is an illusion risk here. When the interface disappears, part of the legibility of who decides disappears with it. Buttons and menus were crude but auditable. An agent that “just handles it” is opaque power unless there are limits, trails, and accountability. Weiser’s and Gates’s prophecy was about usability and diffusion. The agent era forces a second question: usability for whom, and with which brakes.
Five practical consequences
For people who build products, teach, or regulate, the disappearance thesis stops being a book citation and becomes a checklist.
-
Design for intent, not only for screens
The main flow stops being “teach the user to find the button” and becomes “capture goal, constraints, and context with minimal friction.” -
Treat the agent as the product surface
Permissions, memory, tools, and action policies are the new design system. Without them, conversation is a demo; with them, it is a system. -
Separate digital intelligence from physical intelligence
Software scales with GPUs and data. The real world scales with energy, mechanics, safety, and maintenance. Strategies that treat “AI” as one thing miss the bottleneck. -
Make the invisible auditable
If the interface recedes, evidence must remain: what was asked, what was done, with which data, under which policy. Event sourcing, logs, and gates are not engineering nostalgia; they are the caption under the invisible computer. -
Do not confuse horizon with decade
Democracies and firms react to layoffs, energy prices, and polarized feeds long before the “computer that disappears” metaphor becomes everyday comfort. Interface disappearance does not cancel transition politics.
A short formulation
Gates: the computer leaves the center of the desk and becomes the network of life.
AI now: the network leaves the center and becomes intent.
The next step, if the glimpse holds: intent stops being typed and becomes continuous context (voice, vision, agents, robots, environment).
This is not the end of computers.
It is the end of computing as a conscious act for most people, in the same way almost nobody “uses electricity”: they only use the world that electricity (and now intelligence at scale) makes possible.
The information highway was paved. What we are seeing in 2026 is traffic changing character: fewer drivers staring at the dashboard, more destinations spoken aloud, and denser infrastructure under the hood. The computer leaves the conversation. Computing power does not.