AI: Summary
Across the four sessions of April 2026, the Future Text Lab circled a persistent and sharpening question: what, precisely, would compel someone to put on a headset and do serious knowledge work? The month opened with Frode Hegland’s “Best of Both Worlds” presentation on foldable knowledge nodes in visionOS and progressed through a framework distinguishing “core” reading/writing space from “contextual” spatial surrounds, a confrontation with the Engelbart-scale question of whether the group is stuck designing mice without knowing what the mouse is for, and a concluding examination of document ownership, citation infrastructure, and the potential of ePub as a future scholarly container. A new participant, Tim Brookes of the Endangered Alphabets Project, introduced the theme of embodiment and manual skill in relation to digital abstraction. Throughout the month, generative writing — writing to discover rather than to transcribe — served as the conceptual spine connecting flat-screen authoring, spatial interaction, and AI-assisted research.
AI: Main Topic
The primary arc of the month was the design and justification of spatial knowledge environments for thinking — not merely for reading or writing in the conventional sense, but for a distinct cognitive activity that does not yet have settled language or settled tools. Frode presented a framework he called “core and contextual,” arguing that the optimal reading/writing zone is roughly the size of an A4 sheet at arm’s length (derived from ergonomic research on eye, head, and neck movement), and that everything beyond this zone — whether on a large monitor’s margins or in XR room-scale — constitutes contextual space for clustering, relating, and hiding/revealing knowledge objects. He grounded this in Galbraith’s dual-process model of writing, which demonstrates that premature outlining constrains idea generation, and proposed that Author‘s features should serve generative rather than presentational purposes. The month’s discussions repeatedly returned to the tension between this framework’s promise and the reality that no one in the group — including Ken Perlin, as Frode noted — has yet found a compelling everyday reason to work inside a headset beyond virtual display mirroring.
AI: Highlights
Jonathan Finn introduced Zettelkasten as a direct analogue to the knowledge-node concept the group has been developing, noting its rigid card-and-link methodology and its tension between structured and freeform thinking. He observed that he has never found a Zettelkasten implementation whose categories matched how he actually works. Frode responded that the group should attempt to reinvent something like it for spatial thinking.
Peter Dimitrios, in the chat on 6 April, wrote “knowledge walk — like walking through a museum?” — a phrase that captured an emerging metaphor the group kept returning to across subsequent weeks.
Brandel Zachernuk raised the concept of “document integrity” — the idea that a 1-to-1 relationship between a surface and a document is a deeply useful cognitive anchor, and that scrolling and tabbed browsing have eroded it. He argued that if every followed link opened a new window, people would treat pages with more dignity and feel more responsible for managing their attention.
Tom Haymes framed the group’s entire endeavour as living in a “punch card era” of headsets, asking: if the current computing GUI is the equivalent of punch cards, what does the 1968 demo look like for XR? He challenged the group to articulate a destination before refining the vehicle.
Frode reported on 27 April that Apple has offered to send a Vision Pro to his development team — a concrete marker of institutional support for the Author and Knowledge Space development work.
Tim Brookes, joining for the first time on 27 April, asked whether XR-based knowledge tools could serve as a bridge between digital abstraction and embodied skill, noting that younger generations increasingly reject manual trades partly because fine motor skill development has been culturally devalued.
Brandel pushed back firmly on the notion that AI-generated creative output constitutes design, distinguishing between the act of producing a polished-looking artefact and the months-long process of resolving what that artefact should be — a point he illustrated with his experience in Apple marketing where visually perfect pages were routinely discarded because they hadn’t resolved the right question.
Peter Wasilko proposed “peripheral semantic multiplexing” — three copies of the same desktop displayed side by side, where the two flanking copies carry different colour-coded semantic overlays (e.g., text density, date of origin) that would be too cluttered if composited onto the main view.
On 20 April, Brandel stressed the absolute necessity of a deep and overzealous undo queue for spatial environments, referencing Jim Harlan’s concept of “context restoration” and noting that the current cultural climate of surveillance capitalism has made comprehensive activity recording politically unpalatable, even though it is technically essential for making users feel safe modifying their workspace.
Frode noted that Claude‘s typography — its visual formatting of responses — is a reason he prefers it for research, and that even in the age of AI-generated text, the craft of typographic presentation continues to matter for readability and skimming.
AI: Insights
The month surfaced a quiet but significant conceptual shift: the group moved from asking “what should a spatial knowledge node contain?” towards also asking “what should you be able to do to a knowledge node with your second hand?” a repeated references to bimanual interaction — one hand selecting, the other hand pulling, pushing, or twisting to extract different layers of information (highlights, references, history) — reframed the design problem from a content question to an interaction-vocabulary question. This mirrors Engelbart’s original insight that the fluency of interaction, not the content it manipulates, constitutes the medium.
Brandel’s objection to the phrase “getting ideas out of your head” — what he called “notional extraction” — was the sharpest conceptual challenge of the month. He argued that the popular framing of creative tools as mechanisms for externalising pre-formed ideas dangerously misrepresents how ideas actually form. Ideas are resolved through dialog with a medium, not extracted from a mental vault. This has direct design consequences: a spatial thinking tool should be built to support an iterative, ugly, sketching process rather than a clean externalisation of finished thoughts.
A tension emerged between Brandel and Frode on the question of native visionOS development versus web-first development. Brandel maintained that the only computing capabilities that will matter long-term are those that exist on the web, and that native platforms serve primarily as training grounds for what the web will eventually absorb. Frode pushed back by noting that the web currently cannot open or address local files in a way that supports a private knowledge workspace, and that visionOS‘s restrictions on opening external documents represent a real bottleneck. The tension was not resolved but was acknowledged as productive by both.
Tom Haymes’s 1962-versus-1968 framing created an ongoing undercurrent: the group may be over-investing in implementation details (the shape of the mouse) without having articulated a clear destination (what augmented intellect looks like in 2032). Frode accepted the critique but noted that Engelbart himself needed Joe to walk through a concrete system, and that embodied experimentation in the headset produces fundamentally different insights from desk-based imagination.
Jonathan Finn’s observation that composers work simultaneously at macro scale (the duration and instrumentation of a symphony) and micro scale (a few notes, a harmony) without a linear path between them provided a useful analogy for the kind of authoring the group is trying to support — work that is neither top-down outlining nor bottom-up free-writing, but a constant oscillation between structure and fragment.
The April 13 discussion about Masaki Hagino’s “subtitle design” — ePub books displayed as one to three lines of text per phone screen, allowing the reader’s mind to refresh between pages — challenged the assumption that more information density is always better. It introduced the idea that in XR, a furled knowledge node might benefit from being poetically sparse rather than informationally dense.
The discussion of academic publishing on 27 April revealed a deep structural frustration: the group collectively recognised that the ACM and similar institutions sell legitimacy rather than access, that the web has technically solved distribution but not trust, and that the core problems of scholarly communication are social rather than technical. Brandel stated this most directly: technology can provide conduits for solving these problems, but the problems themselves are political, and any framing that treats them as purely technical is misguided.
Tom Haymes’s observation that AI “gives you space” — saving time on clerical work but often making the total task take longer because it enables working at a higher level — reframed AI not as an accelerant but as a cognitive elevator. This parallels the group’s spatial-computing thesis: the value of additional space (whether screen space, room space, or cognitive space) is not efficiency but the opportunity for richer, more resolved thinking.
The month ended with Peter Wasilko proposing that a future document format should use standoff markup — separating raw text from its structural annotations within a single package — and volunteering to write a formal piece on this for the next Future of Text book.