Future Text Lab — 20 April 2026
AI: Summary
This session explored the tension between spatial movement and sedentary interaction in XR knowledge environments, the philosophical gap between incremental implementation and larger augmentation visions, and the nature of externalized knowledge and how it should be manipulated through gesture, voice, and overlay. The group discussed voice commands as an alternative interface for Author on visionOS, the concept of peripheral semantic multiplexing across mirrored displays, the role of AI as a dialogic medium for thought rather than a replacement for creative resolution, and the question of what form externalized knowledge should take when freed from flat screens. A recurring thread was the difference between designing interactions and merely executing them — and whether the community is building toward a clear enough vision of augmented thought for 2032.
The expanded node architecture in Author now permits substantially more content per node in XR, changing the relationship between distance, legibility, and progressive disclosure. When nodes can hold a full screen’s worth of information, their closed state becomes a kind of automatic icon — legible at a distance as shape and layout rather than text. This shifts the function of walking toward a node from navigation to a form of cognitive zooming, where spatial proximity doubles as semantic attention.
The tension between sitting and moving in XR is not a binary design choice but a question of modal switching. The analogy to running shoes — one does not need to leave the house, but the affordance of doing so changes what is possible — reframes the debate. Spatial memory is activated by physical movement through a knowledge space, and this activation is itself a form of thinking, not merely retrieval. The challenge is building interfaces that reward both stillness and movement without punishing either.
Peripheral Semantic Multiplexing proposes using side displays not as additional content surfaces but as alternative lenses on the same content — color-coded overlays revealing facets like text density, provenance, or temporal origin that would clutter the primary workspace if superimposed. This idea connects to a broader principle: that XR’s advantage may lie less in more space and more in simultaneous view specifications of the same information. The observation that web pages opened in parallel can communicate with each other via shared browser context suggests a near-term technical path for prototyping this.
The distinction between 1962 and 1968 in the Engelbart timeline was offered as a provocation: are we still articulating the vision, or are we already demonstrating a system? The counter-argument is that Engelbart’s 1962 paper itself described a walkthrough of a working system, and that the community’s iterative experiments — voice commands, node architecture, overlay concepts — are the equivalent of his component-level demonstrations that built toward the Mother of All Demos. The unresolved question is whether these components are converging toward a coherent whole or merely accumulating.
The framing of “writing is thinking” versus “thinking is writing” revealed a deeper disagreement about whether text-based externalization is the primary cognitive medium or one mode among many. The suggestion that thinking may precede and exceed writing — and that XR environments should support non-textual forms of thought generation — challenges the text-centric assumptions embedded in tools like Author. Generative writing, where the act of composing changes what the author thinks, remains central, but the space beyond it — where gesture, spatial arrangement, and voice contribute to thought formation — is where the community sees unexplored territory.
The use of AI as a conversational partner that expands mental workspace indefinitely was described as qualitatively different from human dialogue, not because it is better but because it removes the constraint of the other person’s time and attention. The resulting problem is convergence: pulling together dozens of semi-related AI conversation threads into a coherent structure is poorly supported by current chatbot interfaces. This is precisely the kind of task a spatial knowledge environment could address — making visible the shape of an ongoing intellectual project distributed across many conversations.
The distinction between design as resolution and design as execution surfaced through the discussion of Claude‘s design tool. A tool that produces polished-looking output quickly is not a design tool if it skips the months of deliberation about why this particular arrangement and not another. The analogy to Apple marketing communications — where every page looked shippable within hours, but the real work was two months of resolving which of the infinite possible pages should exist — applies directly to knowledge work. The externalization of a thought is not the thought; it is one step in a dialogue with the medium that produces the thought.
The idea that a second hand could serve as a gestural modifier in XR — pulling highlights forward, pushing references behind, sliding to hide or reveal — reframes interaction from toolbar-and-button command to something closer to instrumental fluency. The skiing analogy is apt: if the interface requires conscious deliberation at every step, the user falls over cognitively. The goal is an interaction vocabulary so internalized that it becomes invisible, allowing the user’s attention to remain on the knowledge rather than the controls.
Context restoration and deep undo were identified as philosophically essential rather than merely convenient. If users are to feel safe experimenting with spatial arrangements, overlay configurations, and view states, they need confidence that any action is fully reversible — including view actions, not just document edits. The observation that the current cultural moment around surveillance has made comprehensive activity recording politically toxic is a real obstacle to implementing what is technically necessary.
The question of what form externalized knowledge should take remains open. Text, image, sound, abstract shape, 3D model, animation — all are candidates, and likely all will coexist. The deeper issue is that a few keywords almost never suffice to recover a note later; the retrieval problem is as much about the form of storage as about the search mechanism. This suggests that spatial placement itself — the method of loci — may be a retrieval mechanism that complements or replaces keyword search in XR environments.
AI: Resources Mentioned
Peak Human by Johan Norberg — book on the rise and fall of civilizations, discussed as a potential future presentation to the group
Collapse by Jared Diamond — mentioned in connection to Peak Human as covering related themes from the decline side
Guns, Germs, and Steel by Jared Diamond — referenced alongside Collapse
Break the Window: Exploring Spatial Decomposition of Webpages in XR — research paper and project shared via chat: https://btw-xr.github.io/
SurfaceXR: Fusing Smartwatch IMUs and Egocentric Hand Pose for Seamless Surface Interactions — research paper on gesture interaction: https://arxiv.org/abs/2603.19529
Mar Gonzalez-Franco — researcher whose LinkedIn post linked to XR/HCI work: https://www.linkedin.com/posts/margonzalez_spatialcomputing-xr-hci-share-7449427344965996544-Wd8Q
Remembering Doug Lenat and His Quest to Capture the World with Logic — essay by Stephen Wolfram on CYC, Eurisko, and Wolfram Alpha: https://writings.stephenwolfram.com/2023/09/remembering-doug-lenat-1950-2023-and-his-quest-to-capture-the-world-with-logic/
TypingMind — chat interface app mentioned as a higher-level AI chat tool, with a YouTube review shared: https://www.youtube.com/watch?v=R7J3hmlayUE
Elephas — personal knowledge assistant app for Mac, review shared: https://skywork.ai/skypage/en/Elephas-AI-Review-(2025)-The-Ultimate-Personal-Knowledge-Assistant-for-Mac/1974527210251022336
Suno — AI music generation platform: https://suno.com/
Top Gear segment on development of auto control systems — shared as instructive on technology adoption timelines: https://youtu.be/fkwGJzU5B-I
“Writing is Thinking” — video linked in chat: https://youtu.be/pR3Fcnshonc
Doug Engelbart — referenced extensively in the context of his 1962 framework paper, the 1968 demo, and his tool/human capability model
Ted Nelson — referenced in connection to transclusion and the intertwingled nature of text
Jim Harlan, UCSD — mentioned by Brandel Zachernuk in the context of context restoration research
Ken Perlin — mentioned as having said he does not work in the headset despite teaching XR
Bob Horn — referenced for his murals and the prior demo of moving a mural space around the user’s head in XR
Claude — discussed as a tool for code-adjacent design work, document verification, and as the AI of choice for its visual interface and honest feedback
Claude Code — mentioned as useful for coders, with LLM code comprehension also valuable for designer-intermediaries
NotebookLM — referenced as a model for viewing sets of documents and links
AI: Important
No participants addressed Claude directly by name during the live discussion. However, in the broader conversation, it was noted that Claude had been used earlier in the day to verify the quality of Visual-Meta embedding in an Author-produced document and found the embedding “super clean” with only minor line break issues. Claude was also described as being used as a development intermediary — asking whether features are possible in visionOS, receiving code suggestions, and relaying them to the development team. The observation was made that LLMs understanding code is “fantastic for middleman designer people” who are not themselves programmers.