AI: Summary
Across four sessions in April 2026, the Future of Text Lab explored what it means to work with knowledge spatially in XR, moving beyond the established metaphors of flat screens and traditional text. The month opened with Frode Hegland’s “Best of Both Worlds” presentation — the idea that nodes can be furled into compact forms and unfurled into full reading surfaces, merging the advantages of text on a surface with text in space. From there, the group interrogated why anyone would actually put on a headset to think, wrestled with the gulf between grand visions and buildable interactions, debated the web versus native platforms as foundations for spatial knowledge work, and circled toward the question of what document format could carry knowledge into a more open, interoperable future. Tim Brookes joined for the first time, bringing perspectives from endangered alphabets and the embodied dimension of writing. Throughout the month, a recurring tension animated every session: the gap between the conceptual promise of spatial computing and the concrete, unglamorous design decisions required to make it genuinely useful for a single human being trying to think.
AI: Main Topic
The month’s primary arc was the relationship between generative writing — writing as a means of producing new thought rather than recording existing knowledge — and spatial computing as a potential medium for that process. Frode framed this through the concept of “core and contextual” space: the core is approximately an A4 sheet at arm’s length, the ergonomic optimum for focused reading and writing regardless of whether the medium is a laptop or a headset; the contextual is everything surrounding that core, which XR can expand dramatically. The central question the group returned to in every session was what belongs in that contextual space, what form externalized knowledge should take, and how a person should interact with it to support the kind of messy, iterative, generative thinking that produces new understanding rather than merely organising existing facts.
AI: Highlights
Frode shared that Author is now fully available in the App Store and that the development team is rebuilding the entire node architecture, enabling nodes to hold far more content — up to a full screen’s worth — which fundamentally changes what the map view can be.
Frode reported that Apple‘s developer relations team offered to send a headset to his programmers, a concrete sign of institutional support for the project.
Frode submitted a paper to ACM Hypertext during the final session, noting that the writing process itself forced him to contextualize and clarify the spatial work in ways that informal discussion had not.
Brandel Zachernuk mentioned Spatial CSS, a position paper published on WebKit explainers, as the foundational infrastructure work he is pursuing — defining how spatial relationships between web elements can be declared without exposing precise layout or privacy-sensitive data like gaze tracking.
Tom Haymes challenged the group with a framing drawn from Engelbart’s trajectory: “We are in 1962, not 1968. What is our vision for where we want to be in 2032?” He urged the community to articulate a larger goal rather than spending all energy on immediate interaction puzzles.
Tim Brookes, founder of the Endangered Alphabets Project, raised the question of whether spatial computing offers a way to re-embody cognition — making the act of thinking and writing manual and physical again — at a cultural moment when manual skills are in decline.
Jonathan Finn introduced Zettelkasten as a framework worth revisiting and potentially reinventing for spatial knowledge work, noting that its rigid categories never quite fit his own practice but that the underlying principle of atomic, linked knowledge units aligns closely with what the group is designing.
Peter Wasilko proposed “Peripheral Semantic Multiplexing” — three copies of the same desktop with different colour-coded semantic overlays on the two side copies, enabling faceted views of the same information without visual clutter on the primary working surface.
Peter Dimitrios shared several recent HCI research links including “Break the Window” by Mar Gonzalez-Franco and colleagues at Google, exploring spatial decomposition of web pages in XR, and SurfaceXR, fusing smartwatch IMUs with hand pose for surface interactions.
Brandel made a forceful argument about the distinction between design as execution and design as resolution — the iterative process of determining which of an infinitude of possible outputs is the right one — pushing back on the notion that Claude‘s design tool or Midjourney constitute design in any meaningful sense.
Frode described meeting Masaki Hagino of Voyager Japan, who presented “subtitle design” or “haiku design” — ebooks containing only one to three vertical lines of text per page, designed to give the mind space to refresh between pages — and connected this directly to the question of what to put in spatial nodes.
No content was addressed directly to Assistant during any session.
AI: Insights
The most significant conceptual development across the month was the reframing of the headset question from “why would you work in XR?” to the articulation of core versus contextual space as medium-independent categories. By separating the focused reading-writing zone from the surrounding knowledge environment, the group moved past the sterile headset-versus-laptop debate. The core can exist identically in either medium; the differentiator is the contextual, which XR can expand from screen margins to room scale or beyond. This reframing allowed the group to stop defending the headset and start investigating what specifically belongs in the extended space.
Galbraith’s 1999 research on outlining — showing that pre-structuring text reduces generative thinking by turning composition into a fill-in-the-blanks exercise — emerged as a design constraint with real consequences. It means that Author’s outline and overview features must be built to reflect rather than prescribe structure, to be tools for seeing what is there rather than templates for what should be. Frode’s decision to rename the new view mode “overview” rather than “outline” encodes this insight directly into the product.
There is an unresolved tension between Brandel’s web-first position and Frode’s native-platform pragmatism that sharpened across the month. Brandel insists that only things that matter for the web will matter for computing long-term, and that the web’s semantic foundations are essential for the kind of cross-document reasoning spatial computing requires. Frode counters that the web currently cannot give a local application access to the user’s own files without cumbersome workarounds, making it inadequate as a foundation for a personal thinking environment. Neither position is wrong; the tension is productive and points toward a hybrid architecture — perhaps something like an evolved EPUB — that could carry web semantics inside an ownable, portable package.
Brandel’s observation about document integrity surfaced a subtle but important distinction: a traditional monitor provides focus but never context. A browser tab has no guaranteed relationship between its spatial boundaries and its content from one moment to the next. Scrolling further degrades this — Frode noted that “we don’t have web pages, we really have web scrolls.” The implication is that spatial computing’s contribution may not be adding a third dimension so much as restoring the integrity of bounded documents, making each piece of knowledge a discrete spatial object with stable identity.
Tom’s analogy between the current state of XR and the punchcard era was not merely rhetorical. He was pointing to the possibility that the entire metaphor of pages and documents — not just their flat rendering — may be the equivalent of punchcards: a legacy representation that masks a fundamentally different kind of interaction waiting to be discovered. This remained an open challenge through all four sessions.
The discussion of what a knowledge node should show when furled revealed a design philosophy: Peter Dimitrios argued strongly for skeuomorphic visual differentiation — book spines, spiral bindings, glyphs — so that glancing across a space tells you what kind of thing each node is without reading any text. This contrasts with the minimalist rectangle approach and connects to the deeper question of whether spatial knowledge environments should feel more like libraries, workshops, or something unprecedented.
Brandel’s insistence on “overzealous undo” and context restoration as prerequisites for spatial workspaces reflects a deeper principle: people will not experiment with rearranging their knowledge if every action feels irreversible. The reference to Jim Harlan’s context restoration work at UCSD positions this as a research-backed requirement, not a feature request. The irony that surveillance capitalism poisoned the cultural reception of system-wide action recording — making Windows Recall feel threatening rather than empowering — illustrates how social problems constrain technical solutions, a theme Brandel returned to repeatedly.
The month’s final session surfaced an unexpected convergence around document formats. Peter Wasilko’s call for standoff markup — separating text content from its annotation layers — inside a single ownable package is architecturally compatible with Frode’s Visual-Meta approach and with EPUB‘s container model. The group did not reach consensus, but Peter agreed to write a focused piece for the next Future of Text book proposing a new format from first principles, giving the community something concrete to argue with rather than circling abstractly.
Tom’s observation that AI gives you space rather than speed — that working with AI often takes longer because you operate at a higher level of quality and ambition — reframes the productivity narrative. Applied to spatial computing, this suggests that the headset’s value may not be efficiency but elevation: not doing the same work faster, but doing work that was previously impossible because the cognitive overhead of managing many knowledge objects exceeded what flat screens could support.