The passage that got me was the moment the ARPANET team realised that the internal messaging system they’d built would turn out to be far bigger than they expected. They had invented email in 1971 and it would take more than a decade to become mainstream, but the small team already sensed they had stumbled onto something that would reorganise office life.

A lot of world-changing computing seems to begin that way: a small group of technical people build something for themselves, then slowly realise they have also built part of everyone else’s future.

Xerox PARC (Palo Alto Research Centre) was like that too. PARC didn’t become important because Xerox won the computer wars (they didn’t even compete). PARC mattered because it helped invent the grammar of personal computing: the mouse, overlapping windows, networked personal computers, WYSIWYG editing, object-oriented software, the idea that a computer could be a personal medium for thought and work rather than a distant machine you submitted jobs to. The computer they built, the Alto, didn’t become the dominant personal computer, but it helped define what the personal computer would become. The Alto inspired Jobs and Apple building the Mac, and Gates and Microsoft based Windows on the Graphical User Interface (GUI) he saw at PARC.

Small teams building experimental prototypes discover the interaction primitives that become the foundation for future interfaces

PARC was a place where researchers built the kind of computational environment they themselves wanted to inhabit: interactive, personal, visual, networked. They espoused the idea that to build the future, you had to live in it. So they built prototypes that were far from perfect but contained new ideas and primitives. The good ones, like the mouse and the GUI, survived, while others, like document centric computing or end user programming, didn’t. (Ironically, our current era of computing makes the dream of end user programming much more likely, and teams like Lovable and Cursor have revisited the primitives PARC invented).

I think AI is in its PARC phase now. Not the phase where everything has been figured out and neatly packaged for mass adoption, or the phase where one canonical product has emerged. We are earlier than that. We are still discovering the interaction primitives. We are still inventing the basic relationship between people and intelligent systems.

That is why so much of today’s AI software feels both impressive and strangely incomplete. The intelligence is real, but the interface is not. Chat is useful, but chat is not the endpoint. One clue that we are still early is how often AI falls back to primitive interfaces, particularly the resurgence of command lines. New interaction paradigms often begin with basic, low-assumption primitives that enable experimentation before hardening into more complex and polished interfaces. Prompts, CLIs, and chat feel less like a final form and more like lightweight scaffolding around an idea that is still being figured out.

The core job of today’s interfaces then is to surface and make obvious the new primitives of this paradigm. Today they look something like: memory, delegation, tool use, ambient assistance, ongoing sessions, voice as control plane, human review, orchestration across systems, and software that can continue moving work forward while you are somewhere else.

Systems like OpenClaw point toward this pretty clearly. OpenClaw is a self-hosted personal agent that can live inside the interfaces you already use: WhatsApp, Telegram, Slack, Discord or the terminal. It gives you a persistent agent you can reach from anywhere, on your own infrastructure, with memory, tools, and ongoing sessions. It’s so obviously not a final form (it’s janky to say the least) but is also so malleable that users can explore how something like this might fit into their daily life. It hints at the shape of the next interface: not a place you go to operate software, but a system you can delegate work to.

The dream of computing has always been to extend human ability

If you go back far enough, this was always the more interesting ambition for computing. Licklider’s idea of man-computer symbiosis was not about replacing people with machines. Engelbart’s vision of augmenting human intellect was not about handing cognition over entirely. The point was to create systems that extended human thought and action. The computer as partner, amplifier, medium.

PARC answered that question for the graphical era. Files, folders, icons, windows, menus, dragging, dropping, pointing, clicking: the desktop metaphor turned computing into a space you could inhabit and manipulate directly. For half a century, most digital work has been built on the assumption that the human sits at the centre of the loop, moving through software one explicit action at a time.

PARC-style computing made the computer easier to operate. Agentic computing makes it easier to orchestrate.

The thing we call a computer has changed before. The first Computers were people, then they were giant machines for calculation, then they became personal workspaces. What is emerging now looks like another redefinition: the computer as a remote environment for cognition and action.

As this model emerges, a lot of what we currently think of as “computer work” will start to look like leftover behaviour from an older era: endless app-switching, email checking, dashboard refreshing and the general sense that work only happens when your face is directly in front of the glowing rectangle. It could be that people will spend less time “on the computer” in the old sense and more time meeting colleagues, customers or users.

But separating the person from the work creates new challenges. The desktop metaphor made causality legible: you clicked, something happened, you could see it. Orchestrated, ambient, asynchronous work is murkier. If the machine continues advancing your work while you’re somewhere else, the question of understanding what it did becomes genuinely hard. That might be the most important primitive still left to invent, a new way of knowing what happened and why.

The rectangle is not going away, but its monopoly on our attention is fracturing, creating room for new, more humane, interfaces.