Can generative AI reduce overhead work that adds no value?

Ida Larsen-Ledet's CHI 2026 paper proposes a distinction between "productive" and "mundane" articulation in people's work with generative AI, arguing that "generative AI will not become a true asset in the workplace unless it allows people to focus on productive articulation that helps them reflect on and refine the objectives of their work, by reducing the need for mundane articulation that makes up the rote work of setting up fundamental context." She defines the mundane as "that which adds no value beyond ensuring that outputs are suitable for the person and the situation," and the productive as "that which progresses not only the generation of output but also how the work is conceptualized" — explaining a problem to a coworker, or "a rubber duck," can itself surface a solution; restating that an "article" means an academic one, not a news one, is pure overhead.

Her mechanism draws on CSCW's notion of articulation work: collaborators maintain "common ground" and divide labor, coordination that is "not by definition good or bad," just part of working together. GenAI imposes an analogous burden — the user must "figure out both what to tell the system and how to tell it" — but because most tools rely on "a superficial notion of context" (organizational or physical location, which app the AI "lives" in, or prior chat turns) rather than genuine contextualization, people keep re-articulating things an "'intelligent' agent" should infer. She frames this within organizational knowledge infrastructures, where genAI is meant to move knowledge from "passive stores" to "active surfacing," yet current tools, "restricted to conversational interfaces" or siloed to one app, "does not match how people work."

This gives a name to a cost nearby design-space papers describe without naming. Do generated analysis UIs really work better than chat? finds non-programmer analysts who "struggled to specify widgets at the right granularity" and had to "think like a software engineer" — a concrete instance of mundane articulation, imposed through a GUI rather than text. Do generated interfaces outperform text-based chat for most tasks? shows people preferring structured interfaces to "conversational chat," but Larsen-Ledet's framing implies modality is not the real lever: a generated UI that can't anchor to context still forces mundane specification, just through widgets. Her closing point — that "the way to create more relevant knowledge experiences at work can be better design, with or without (better) AI" — echoes Can AI boost how teams work together?'s case that realizing AI's workplace value is a design problem, not a model-capability one.

The paper is explicitly conceptual. Larsen-Ledet states her purpose is "not to provide a concrete list of what is mundane and what is productive" but to "propose a direction of inquiry," and she leaves open whether the line is even stable — stating a job title might be mundane in one situation and load-bearing in another — or whether mundane and productive are discrete categories or ends of a spectrum. No system, user study, or measurement backs the distinction; it is offered as a lens for future design and evaluation. What follows at that strength is narrow: the distinction usefully names a cost that other work documents empirically, but operationalizing "mundane" versus "productive" in any specific tool is left for someone else to establish.