Static plant.
Immense complexity, but the room itself is stationary. State is high-stakes; the rate of change is slow and the operator's frame of reference is fixed.
The Receding User
The user hasleft the room.
The interface still has work to do.
A long-form essay on what happens when AI agents act while we're not watching.
01 · The Arc
Sixty years of shrinking distance.
In 1963, Ivan Sutherland's Sketchpad let a user draw a line on a screen and have the line behave as a line — snapping, staying parallel, holding length. The drawing answered back.
Five years later, Douglas Engelbart's NLS demonstration added the mouse, hypertext, structured outlines, multi-window editing, and live collaboration with a remote colleague.
Together, they established something computing had not contained before: the user and the digital object could share a surface, in real time, in the same act.
Co-presence became what an interface was for.
Over the next four decades — the Macintosh (1984), Photoshop (1990), the multiplayer web, touch, Figma — the distance between intent and visible consequence got smaller at every step. By 2022, “intuitive” meant “closer to the surface.”
02 · The Break
In November 2022, that arc broke.November 2022. The arc broke.
The pattern that has stabilized across ChatGPT and what followed is, structurally, batch computing with natural language input. The user types a prompt; the system thinks for some seconds, or minutes, or thirty minutes; the user reads the result.
The user is elsewhere while the system works.
The new distance runs through time.
The user remains in the room and on the surface.
What they no longer share is the moment of action.
03 · What Actually Changed
Three shifts happened at once. They keep getting conflated.
Each one is something you can handle on its own. The three together is what's actually changing the shape of working with a machine. And most of the interfaces we have today are still working on one of them at a time. Usually input.
04 · The Pattern
Review · Confirm · Undo
The dominant interface pattern today — diff views in coding agents, accept/reject buttons in writing tools, regenerate buttons under every chat response — is version control applied to model output.
It is useful. Its limit is structural.
USER ABSENT
0.5 secA pause within an action
30 secA turn in a conversation
30 minAn episode of its own
You cannot review one hundred things.
Review / confirm / undo gets you through a couple rounds of agent work. Once the agent has been off doing things for half an hour, you need something else. That something else hasn't really been figured out yet.
05 · The Precedents
The interface for the absent user already exists.
It just was not the part of computing that consumer software was modelled on. Four fields offer well-developed answers to versions of this question.
Air Traffic Control
Glanceable state over event narration. The controller's competence is reading the whole airspace at once, intervening only at exceptions.
SCADA / Process Control
The failure mode is alarm flooding — too many alerts fire to act on any of them. Operators read trends and deviations; individual events stay subordinate.
The Unix Tradition
Daemons, cron jobs, ps, top, tail -f, system logs. The forms current AI products need have been sitting in plain view since the 1970s.
Spreadsheets
A
B
C
12
8
20
C1 = A1 + B1
Declare the relationships once. The system maintains them while you are elsewhere. The simplest case of coherence held on the user's behalf.
None of this needs reinventing. It needs translating — into the visual vocabulary of consumer software, and into the rhythms of agentic interaction.
06 · The Metaphor FamilyNuclear Control Room NASA Ground Control Starship Bridge Aircraft Cockpit
Why cockpit, specifically?
The cockpit sits in a family — NASA Mission Control, nuclear power plant control rooms, the bridge of a sci-fi starship. All four are interfaces for operators who cannot directly touch the thing they are operating.
The cockpit is the strongest of them — because pilot and plane are co-located in something that is moving. The interface has to surface velocity and trajectory alongside state. The agentic situation has that shape exactly.
Remote vehicle.
Tracks a moving system from a fixed ground station. Communication latency is built in. Closer to cockpit, still one step removed.
Cultural ideal.
Sci-fi version of the same intuition: a crew reading state, trajectory, and exception from a shared persistent surface. Why the metaphor lands fast.
Moving feedback.
Pilot and plane co-located in a continuously changing system. The interface has to surface velocity and trajectory alongside state. This is the agentic shape.
Agentic systems are a moving-feedback problem. That's the cockpit family.
07 · The Destination
Build the cockpit.
A persistent surface. State, history, ongoing processes. Deviations from expectation. Clear points of intervention. The chat input becomes one panel among several.
This is the shape of the interface for the receding user: a place to arrive at, where the system has held coherence in your absence. Well-designed reading is the deliverable.
Bret Victor, in Magic Ink (2006), drew the distinction: most software is treated as manipulation software, in which the user acts continuously. Most software should be information software, in which the user reads, and the system's task is to design the page well.
When the user is absent during the action, what they need on their return is information software.
08 · Coda
Sixty years ago, Sutherland made the drawing answer back.
The next sixty are about what the system shows us when we return — and how it holds coherence in our absence.
That is the interface design problem the next decade will answer. Or fail to.
Written by Jonny Dubowsky. Synthesized from The Receding User: Designing for Absence by Saleh, Interface Studies (2026). → watch the original
This site is itself an argument about interface design. If it worked, you noticed.
09 · References
Sources.
Every citation in this essay, with a link. We owe the reader at least this much.
Sutherland, I. E. (1963). Sketchpad: A Man-Machine Graphical Communication System. MIT PhD thesis.
https://dspace.mit.edu/entities/publication/b5e8025c-c8b2-4843-84e0-76db824e07e6Engelbart, D. C. (1962). Augmenting Human Intellect: A Conceptual Framework. SRI Summary Report.
https://www.dougengelbart.org/content/view/138/Engelbart, D. C., & English, W. K. (1968). A research center for augmenting human intellect (the “Mother of All Demos”).
https://www.dougengelbart.org/content/view/374/Shneiderman, B. (1983). Direct manipulation: A step beyond programming languages. Computer, 16(8), 57–69.
https://ieeexplore.ieee.org/document/1654471Hutchins, E. L., Hollan, J. D., & Norman, D. A. (1985). Direct manipulation interfaces. Human–Computer Interaction, 1(4), 311–338.
https://www.tandfonline.com/doi/abs/10.1207/s15327051hci0104_2Weiser, M. (1991). The computer for the 21st century. Scientific American, 265(3), 94–104.
https://www.scientificamerican.com/article/the-computer-for-the-21st-century/Bentley, R., Hughes, J. A., Randall, D., Rodden, T., Sawyer, P., Shapiro, D., & Sommerville, I. (1992). Ethnographically-informed systems design for air traffic control. CSCW ’92.
https://dl.acm.org/doi/10.1145/143457.143470Hutchins, E. (1995). Cognition in the Wild. MIT Press.
https://mitpress.mit.edu/9780262581462/cognition-in-the-wild/Rasmussen, J. (1983). Skills, rules, and knowledge; signals, signs, and symbols, and other distinctions in human performance models. IEEE Transactions on Systems, Man, and Cybernetics, SMC-13(3), 257–266.
https://ieeexplore.ieee.org/document/6313160Vicente, K. J. (1999). Cognitive Work Analysis: Toward Safe, Productive, and Healthy Computer-Based Work. Lawrence Erlbaum Associates.
https://www.routledge.com/Cognitive-Work-Analysis-Toward-Safe-Productive-and-Healthy-Computer-Based-Work/Vicente/p/book/9780805823974Nardi, B. A. (1993). A Small Matter of Programming: Perspectives on End User Computing. MIT Press.
https://mitpress.mit.edu/9780262140539/a-small-matter-of-programming/Victor, B. (2006). Magic Ink: Information Software and the Graphical Interface.
https://worrydream.com/MagicInk/Baranes, A., & Marchant, R. (2026). Shaping the future of AI interaction by reimagining the mouse pointer. Google DeepMind blog, 12 May 2026.
https://deepmind.google/blog/ai-pointer/Clicky (2026). AI cursor companion for macOS, by Farza.
https://www.clicky.so/Flath, I. (2026). How Clicky’s AI Interface Works.
https://isaacflath.com/writing/how-clicky-worksSaleh (2026). The Receding User: Designing for Absence. Interface Studies, YouTube. The source essay this site synthesizes; watch it in full.
https://youtu.be/C0aCBq8SlYI