Generative Writing in Core and Contextual Environments
We looked at generative writing, where writing does not just record pre-formed thoughts but generates new understanding through the act of writing itself, in Core & Contextual environments.
AI: Summary
This session wove together three interconnected threads: the design philosophy behind generative writing tools (specifically Author), the emerging framework of “core and contextual” writing spaces as a bridge between flat displays and XR, and the challenge of what meaningful interaction in XR actually looks like for knowledge work. The conversation moved from AI’s role in education and writing practice, through the cognitive science of outlining, to a provocative closing framing: that current XR interfaces are the punch cards of their era, and the question before this community is what the equivalent of Engelbart’s 1962–1968 project would look like today.
A key insight emerged around outlining as a cognitive straitjacket: research by Galbraith suggests that outlining before writing reduces knowledge generation because it turns writing into a fill-in-the-dots exercise rather than a generative act. This has direct design implications — an outline feature in a writing tool can work against the very thinking the tool is meant to support. The response to this tension is the proposed distinction between ‘outline’ and ‘overview,’ where the latter allows a non-prescriptive view of the document without imposing structure in advance.
The ‘core and contextual’ framework was articulated as a media-neutral concept, not a hardware distinction. The optimal reading space (roughly an A4 sheet at arm’s length, ~60cm) is grounded in human ergonomics and visual field constraints — not in paper culture. The insight is that headsets subsume flat displays rather than replace them: a virtual display in a headset can be functionally identical to a laptop screen. What changes is the contextual space available beyond that core, which expands significantly when the user moves.
A productive reframing emerged around peripheral/context and core: it was proposed that the peripheral contextual space should inform the core writing space, not merely extend from it. The analogy offered was a grocery store — peripheral items suggest and prompt the recipe being composed at the center.
A significant tension was identified around what to actually show in XR contextual space. Connections drawn as lines between nodes were found, in practice, to feel physically intrusive and space-consuming — like washing wires. This suggests that spatial topology and proximity may need to carry semantic weight without explicit visual connectors. Color was proposed as a filtering mechanism, but quickly refined: rather than using color for categories (limited to ~5–7 distinguishable hues), color or visual filters could encode dynamic prioritization driven by AI inference about what the writer is currently working on.
The interaction design gap in XR was identified as the loss of the keyboard and trackpad when moving through space, and the corresponding gain of the whole body. A game interaction — where different physical locations on the body map to different tools — was cited as an underexplored model for embodied computing.
The ‘pinch and move the world’ interaction (attributed to Brandel Zachernuk from the Bob Horne mural project) was highlighted as a breakthrough: rather than navigating through a fixed virtual space, the user pulls the world around themselves, avoiding locomotion sickness while preserving spatial agency.
A distinction was drawn between two tiers of spatial relationship: tight clusters of objects (analogous to what’s inside a window) and diffuse associations between clusters (analogous to relations between windows). The proposal is that the latter might be expressible declaratively — like CSS — rather than requiring explicit pixel-level authorship.
Privacy was identified as a structural constraint on XR experimentation, not a philosophical concern. The specific example: gaze data in visionOS is architecturally withheld from developers to prevent exploitation, which limits experimentation but is treated as a non-negotiable constraint worth designing around.
The closing provocation reframed the entire session: if current XR interaction paradigms are punch cards, what are the human limitations of today’s GUI environment that an ambitious HCI research project — in the spirit of Engelbart — would set out to overcome in the next five years?
A smaller but notable insight: AI assistants differ significantly in their editorial behavior — one major model (Google Gemini) was observed to give uncritical positive feedback on writing, while Claude was described as considerably more rigorous in identifying structural problems.
Author — Frode Hegland’s writing application, under active development, demonstrated live during the session.
Apple Vision Pro / visionOS — discussed extensively as the primary XR development platform in the group’s context.
webXR — the development environment used by Brandel Zachernuk; noted as a place for experimentation due to its open affordances, despite limitations in text rendering.
NotebookLM (Google) — mentioned as a tool for building searchable libraries of sources and finding contextual connections across readings.
Mozilla Hubs — referenced as an early XR environment used for early spatial experiments with the group.
Bryce 3D and Metacreations Poser — older 3D applications referenced as examples of tools with tangible, discoverable handles for interaction.
Photoshop, Maya, Blender — referenced in the context of how orienting oneself in relation to work is as important as the work itself.
LinkedIn / Chrome — cited in a discussion of browser fingerprinting and the privacy risks of spatial web data; LinkedInspecifically mentioned for tracking browser plugin lists to infer user behaviour.
Spatial CSS — a position paper / WebKit explainer authored by Brandel Zachernuk, describing how to conceptualize space behind and in front of a window. Available on the WebKit explainers site (exact URL not given in session).
Ministry for the Future by Kim Stanley Robinson — mentioned as an example of mosaic/scrapbook writing where different text types coexist at the same point size, raising questions about how XR could represent the relationships between such fragments.
Moby Dick and Les Misérables — referenced alongside Ministry for the Future as canonical examples of patchwork or multi-register literary texts.
David Galbraith — researcher whose studies on outlining and knowledge-constituting writing were cited; noted as active circa 1999 and relevant to the cognitive science of writing.
Doug Engelbart — referenced as the historical anchor for the session’s closing provocation; specifically his 1962–1968 project to rethink human-computer interaction from first principles.
Alan Kay — referenced in the context of the conceptual invention of windows as a computing paradigm.
Bowen and Watkins — cited for the phrase “naive intern” as a characterisation of current AI assistants.
Dave Millard — Frode’s former professor; credited with the insight that a map constrained to a plane gives meaningful value to what is in front of and behind it, leading to the snap-to-Z-equals-zero feature in Author.
Carl Linnaeus — referenced as a historical figure whose slip-based knowledge organisation inspired early spatial experiments in Mozilla Hubs.
Travis Kalanick (Uber) — cited as an example of a maximally extractive tech actor in a discussion of privacy risks.