---
title: "Spaces: The Geography of Computing"
code: "RP-0128"
language: "en"
canonical: "https://regentspark.ai/RP-0128/"
html: "https://regentspark.ai/RP-0128/"
markdown: "https://regentspark.ai/RP-0128.md"
updated: "17 April 2026"
---
# Spaces: The Geography of Computing
*This essay follows Spaces: how do we hold work together across tools?. That piece argued that work needs a shared substrate beneath changing tools. This one asks what would make that world legible to a person trying to orient, return, and remember.*
> **Summary:** Spaces are not only useful because they scope work. They are useful because humans remember places better than paths. A real computing substrate must therefore support not just containment, but orientation, landmarks, scale, traces, and return.
## We remember places better than paths
> **Summary:** Humans are unusually good at remembering place: location, scale, landmarks, and routes. Computers mostly make us remember names, apps, and file paths instead.
### Why Google Earth feels easier to navigate than a computer
It is often hard to find something on a computer, even when we know it is there somewhere. But in Google Earth, we can start from the whole planet and steer continuously toward a country, a city, a neighborhood, a street, and finally a house in one unbroken movement.
That matters because the movement itself is doing cognitive work.
We are not just recalling a label. We are:
- orienting ourselves
- recognizing landmarks
- narrowing scale
- using proximity
- steering toward remembered regions
The route becomes part of the memory.
This is also why the memory-palace idea is so powerful. A memory palace is not magic. It works because spatial arrangement becomes recall structure. Put an idea in a place and the place helps you retrieve the idea later.
The classical method of loci is one of the oldest explicit demonstrations that memory and place are deeply entangled. See Britannica's overview of the method of loci β the old mnemonic trick of placing ideas in imagined rooms and then walking back through them.
### Why spatial memory works without immersive 3D
There are really two powers here:
- **local spatial memory** inside a place
- **large-scale navigation** between places
The first gives us the memory palace. The second gives us the map.
The important clarification is that none of this requires immersive 3D. The useful part is not virtual reality. It is continuity of orientation across scales. A flat map can do that perfectly well.
> **Alt text:** A six-panel zoom sequence shows Europe, France, a city block map, a building facade, a floor plan, and finally a room and desk with notebook and glasses.
> **Description:** The illustration is a horizontal sequence of six connected panels across the page, reading from left to right as a zoom from larger geography to intimate workspace. On the far left is a pale map of Europe with a dark dot, next to a larger country-shaped map labeled by context as France, with a dark center point; thin curved lines link these scales into the next panel. In the middle is a detailed street-map block, then a narrow view of a multi-story building facade, then a simple black-outline floor plan with one room highlighted in a light tan fill and a small arrow pointing upward beneath it. The sequence continues into a furnished room with a bookshelf, window, and table, and ends on the far right with a close-up desk scene showing an open notebook, eyeglasses, and a pen, with another blurred windowed building view behind it. The image communicates continuous navigation across scalesβfrom continent to country to city block to building to room to deskβso that a person can maintain orientation while zooming in.
> **Image source:** content/RP-0128/illustrations/spaces-geography-zoom.png
```mermaid
flowchart TD
World[π World] --> City[π City] --> Building[π Building] --> Room[π Room] --> Desk[π Desk] --> Object[π Thing in view]
Person((π€ Person)) -->|steers by scale and landmarks| World
```
Kevin Lynch's The Image of the City (1960) is still the clearest argument that paths, landmarks, edges, districts, and nodes are not decorative urban features. They are the structure people use to stay oriented.
## Today's software gives us containers, not geography
> **Summary:** Current computing gives us containment without geography. When the system does not support orientation, people fall back to search instead.
### Folders hold things without telling us where we are
A folder is a weak kind of place. It contains things, but it does not really give us geography.
It does not tell us:
- what is nearby
- what scale we are at
- how this place relates to a broader territory
- what the landmarks are
- how to steer toward something rather than search for it by name
### Apps create local worlds that do not connect to each other
Apps are often worse. Each app is its own tiny world with its own rules, but those worlds do not compose into a larger geography. We open a tab, click a link, switch a window, and get teleported into another silo with almost no sense of traversal.
Some tools recover one slice of the missing model. Figma preserves orientation while you zoom and pan. Muse and tldraw let arrangement hold thought on a canvas. Maps prove that very large territories can stay navigable when scale changes remain continuous.
Virtual worlds point in the same direction at a larger scale. Second Life, Active Worlds, and sandbox cities such as Grand Theft Auto teach neighborhoods, meeting points, and routes when a world stays stable enough to inhabit.
But these wins stay local to single products. The geography disappears as soon as you leave the app.
The lesson is not that everything should become a metaverse. It is that orientation gets easier when the system behaves like a place instead of a stack of jumps.
### Search became the patch for broken navigation
That is why so much modern computing quietly depends on search.
Search is useful, but it should be a fallback, not the primary way to recover orientation. When users stop navigating and start searching for everything, that usually means the environment has failed to become legible as a place.
```mermaid
flowchart LR
Physical[π Physical world
landmarks Β· routes Β· scale] --> Recall[π§ Spatial recall]
Digital[π Current computing
apps Β· tabs Β· paths Β· jumps] --> Search[π Search fallback]
```
## A space becomes useful when it becomes a place
> **Summary:** Human environments already come with meaningful scales, and a digital world becomes useful when spaces turn into places that support orientation, memory, and return.
### Room, house, street, city, and world already mean different things to people
Different spatial scales already carry different expectations for humans.
- **Room:** Immediate activity. What is near me right now. What is on the desk. What is in reach.
- **House:** Belonging, ownership, stable context, repeated return. This is where I live. This is my base.
- **Street or block:** Adjacency, neighbors, nearby services, visible alternatives, movement between related places.
- **City:** Districts, infrastructure, public systems, transportation, density, discoverability.
- **World:** The atlas view. Exploration, routing, search, distant travel, broad context.
Computing today flattens many of these scales into one mess. A folder, a workspace, an app, a browser tab, a dashboard, and an account are all different structures, but they rarely form a coherent scale ladder.
A real computing geography would let someone:
- zoom out to see the wider territory
- zoom in to recover local detail
- move laterally between neighboring spaces
- keep orientation while changing scale
- understand whether they are in a room, a house, or a city
This also suggests a useful distinction:
- **space** is the general primitive
- **environment** may just be a special kind of root or host space
That keeps the ontology cleaner. We do not need separate nouns for every level. We need one spatial primitive that can carry different roles at different scales.
> **Alt text:** Split illustration showing a navigable city map on the left and a laptop with scattered app windows, folders, and messages on the right.
> **Description:** The illustration is split into two main halves: on the left is a stylized city map with pale streets, blocks, parks, waterways, and a river or canal cutting through the lower middle, while on the right is a laptop surrounded by floating software windows, folders, chat bubbles, and file cards. The left side includes recognizable civic landmarks and symbols embedded in the map, such as a clock tower near the upper right, a domed building near the center-right, and a station-like icon and bridge elements near the lower center, all reinforcing the idea of navigable geography. The map reads as a coherent place with neighborhoods, routes, and destinations connected by roads, whereas the right side reads as fragmented digital interfaces layered above the laptop screen, suggesting apps and files rather than a continuous environment. The laptop sits in the lower right foreground with a hand on the trackpad, while the windows and document panels hover above it in staggered layers, implying many disconnected tasks or contexts. Together, the image contrasts legible spatial orientation with the cluttered, non-geographic structure of current computing, matching the articleβs argument that software should behave more like a place than a stack of screens.
> **Image source:** content/RP-0128/illustrations/spaces-fragmented-laptop.png
```mermaid
flowchart TD
World[π World] --> District[π District] --> Room[π Room] --> Desk[π Desk]
Room --> Person((π€ Person))
Desk --> Object[π Object at hand]
Desk --> Memory[(π§ Trace of what happened here)]
```
### A place is a space with memory, objects, people, and traces
This is where the four primitives start composing more clearly.
The cleanest version is simple: spaces plus memory produce place.
A space without memory is just a container.
A space without objects is just an empty shell.
A space without people is just a model.
What makes a place feel like a place is continuity.
You return and the traces are still there. The same objects still matter. The same people still have history with one another. What happened here remains legible enough to orient you.
That is why Why does your computer remember everything but understand nothing? matters so much to the spaces argument. Memory is not an adjacent feature. It is what turns a room from a container into a place.
```mermaid
flowchart TD
Person((π€ People)) --> Place[π Place]
Space[π Space] --> Place
Object[π Objects] --> Place
Memory[(π§ Memory)] --> Place
Place --> Geography[π Geography across places]
```
Yi-Fu Tuan's Space and Place is a good philosophical companion here. His basic distinction is still useful: place is not just bounded space, but space that has become meaningful through experience, attachment, and return.
### Interfaces should preserve orientation while we move
This section does not try to specify one renderer. It tries to identify the minimum experiential guarantees.
- **You are always somewhere.** The system should always make it clear what place you are in, not just which app is open.
- **Moving should preserve orientation.** Changing level, entering a subspace, or returning to a broader map should feel like traversal, not teleportation.
- **Scale should mean something.** Deeper should usually mean more specific. Wider should usually mean broader. The gradient cannot invert randomly.
- **Landmarks should stay stable.** People build cognitive maps around fixed points. If the landmarks keep moving, users fall back to search because navigation stops being trustworthy.
- **Arrangement should carry memory.** Inside a place, local layout should help recall. Spatial arrangement is part of the memory system.
- **Returning to a place should reveal traces.** If something happened here, evidence of that should still be discoverable there.
- **Geography should beat hierarchy when possible.** Pure trees are too weak. Real places support containment, adjacency, districts, routes, shortcuts, and landmarks all at once.
- **Search should supplement place, not replace it.** A healthy spatial system reduces dependence on search by making location legible again.
- **Multiple lenses should render the same world.** The same underlying world should survive being viewed as a map, a dossier, a list, a canvas, or an agent-oriented surface.
## The computer needs geography in the substrate
> **Summary:** Many tools have rediscovered local geographies, but they stop at the app boundary. The missing piece is a shared lower layer that lets different interfaces preserve the same world.
### Maps, canvases, worlds, and the web each prove part of the model
A few families of systems have already proved parts of this model.
**Maps and zoomable interfaces.** Maps solve scale and orientation unusually well. They let someone move from global context to local detail without losing their place, as long as landmarks remain legible and the transition stays continuous. Zoomable interfaces inherit that strength when zoom means scope rather than just bigger pixels.
Tolman's classic paper Cognitive Maps in Rats and Men (1948) is the old root of this argument. The mind does not merely remember turns. It builds a map. Later path-integration research, such as Etienne and colleagues' review on path integration in mammals, makes the same point from a different angle: orientation depends on continuous movement and stable reference points.
Figma's own help docs on panning and zooming in FigJam are banal on the surface, but the product lesson is deeper: users can maintain orientation because movement across scale is continuous rather than teleportational.
**Spatial canvases.** Spatial canvases show that arrangement itself can carry thought. Muse, tldraw, and similar tools let proximity, grouping, and return do part of the memory work. What they usually lack is a durable world beneath the view, so their geography stays trapped inside one tool.
Adam Wiggins' Muse retrospective is useful here because it shows what a spatial canvas can do for thought without claiming that the whole computer should become one endless whiteboard. The official tldraw canvas interaction docs make the same point from the product side: pan, zoom, and viewport control are part of the cognitive substrate of the canvas, not just cosmetic polish.
**Virtual worlds and world engines.** Virtual worlds and game environments show the larger-scale version. Second Life, Active Worlds, and sandbox cities such as Grand Theft Auto teach districts, routes, meeting points, and repeated return. The important lesson is not spectacle or VR. It is that orientation becomes easier when a system behaves like an inhabitable place instead of a stack of jumps.
The History of Second Life is useful because it shows the world evolving into a resident-built geography of named regions, telehubs, land, and community gathering spaces rather than just a graphical chat room. Active Worlds carried a similar intuition with personal worlds, public building worlds, and a persistent community.
**Operating systems and the web.** Filesystems give containment. Operating systems give bounded environments. The web gives global addressability and loose interconnection. Each proves one useful property. None of them, by itself, gives us a shared semantic world where people, spaces, objects, and memory remain legible across tools.
Each family proves something real. The recurring failure is that the geography stays local, app-owned, or renderer-dependent rather than becoming a general property of the system.
### Different interfaces should render the same underlying world
The problem is that most of these geographies stop at the app boundary.
They can produce a compelling local world, but they do not provide a shared world beneath the rest of computing. Close the shell, leave the app, or break the integration and the geography disappears. The place was only a presentation, not part of the underlying world.
That is where this topic connects directly to the super-app trap. A super app can simulate geography inside its own walls. What it cannot do is make geography a general property of the system.
The question, then, is not which spatial UI looks nicest.
It is:
**What lower-level guarantees would let many different interfaces preserve the same geography of computing?**
> **Alt text:** Overhead illustration of a map, a collaborative table with papers and people, a folder stack, and a checklist board arranged as a model of digital space and place.
> **Description:** The illustration is arranged like a flat overhead scene with several separated panels around a central working area. At the top center there is a framed map with abstract land-colored regions and thin route lines, suggesting a broader territory or world view. In the middle, a large rectangular wooden table is surrounded by several small people seen from above, each handling papers, notebooks, and folders; the table also holds a coffee cup, a small tool or pen, and stacked documents, communicating collaborative work in a shared place. On the left is a large blank sheet or canvas with a few broad paint-like strokes, while on the right is a tall file stack or folder tray packed with layered documents and tabs, contrasting open creative space with compressed storage. Along the bottom is a wide checklist-like board with rows of lines and bullet/check marks, acting as a summary or planning surface beneath the scene. Together the image communicates the articleβs idea that computing should be legible as a place with scale, landmarks, objects, and traces rather than a set of disconnected files or apps.
> **Image source:** content/RP-0128/illustrations/spaces-many-views.png
```mermaid
flowchart TD
Substrate[π Shared substrate] --> Map[πͺ Map view]
Substrate --> Dossier[πͺ Dossier]
Substrate --> List[πͺ List]
Substrate --> Canvas[πͺ Canvas]
Substrate --> Agent[πͺ Agent context]
```
### What the substrate would need to guarantee
This is the deeper systems question sitting underneath the essay. Not a full specification, just the minimum contract.
A real computing geography would need at least this much:
- **Persistent spaces across scales.** The same primitive must survive as room, house, district, city, or world.
- **Stable identity.** People, spaces, and objects must remain themselves across devices, views, and neighboring context.
- **Containment plus adjacency.** Geography needs both inside and next to, not just a tree.
- **Meaningful traversal.** Entering, leaving, zooming, and lateral movement should have consistent semantics instead of arbitrary jumps.
- **Traces in place.** Memory, provenance, and evidence of what happened should remain recoverable where it happened.
- **Many views over one world.** A map, list, dossier, canvas, or agent view should be a projection of the same underlying place, not a new silo.
The point is not to make software look more spatial. The point is to let digital work inherit the cognitive advantages of place: orientation, return, legibility, and memory.
A folder gives containment. An app gives function. A search bar gives retrieval. None of those, by themselves, gives you place.
Spaces: how do we hold work together across tools? argued that work should survive crossing tools. This essay adds the next demand: once it does, a human should still be able to find their way inside it.
That is what physical space gives us almost for free. Computers have still barely learned to do it.
That is why geography cannot stay trapped inside individual apps. It has to live in the substrate.
*Related: Spaces: how do we hold work together across tools? Β· Why do I get lost on my computer? Β· Why does your computer remember everything but understand nothing? Β· Objects: why are our things trapped inside apps?*