Can Human Factors design fixes restore GenAI productivity losses?

Building on their own diagnosis that GenAI reproduces automation's "ironies" (see Does generative AI productivity actually improve or decline in practice?), Auste Simkute, Lev Tankelevitch and colleagues propose five design responses drawn from decades of Human Factors research on human-automation interaction: "providing continuous relevant feedback to users," "enabling system personalization," "applying ecological interface design," "using task stabilization and interruption timing techniques," and "enabling clear task allocation between users and systems."

They state the three goals these five approaches share: to "increase user agency in how they adapt the GenAI support to users' preferred ways of working," to "increase users' situational awareness of system changes and potential errors," and to "increase user flexibility through the more granular application of AI support to their tasks, freeing users from having to make a binary decision of either using GenAI tools potentially ineffectively or not using them at all." They develop continuous feedback as the primary example: during monitoring, feedback should tell users "if their requests have been received," "if the actions of the automation system are being performed properly," and "if any errors are occurring," and should be "informative but non-intrusive," adapting "to the interactive style of the participants and the nature of the problem." Feedback is also tied to explainability, so users can "build mental models of the system's behaviour" and anticipate where it will fail.

The proposal shares Larsen-Ledet's framing that fixing GenAI's workplace costs is a design problem rather than a model-capability one — her note argues "the way to create more relevant knowledge experiences at work can be better design, with or without (better) AI," and "clear task allocation" here is this paper's version of the same claim, applied to deciding which parts of a task the system versus the user should own. It is the design-response counterpart to the role-shift and workflow-restructuring costs Does generative AI productivity actually improve or decline in practice? documents, aimed directly at the same mechanisms. It also parallels How should generative UI be organized as a design space?'s project of sorting GenAI interaction choices into named dimensions, here applied to ecological interface design and task allocation rather than audience and constancy.

The authors present these five approaches "throughout… focus[ed] on the programming domain as an example," not as tested interventions: the excerpt gives no evaluation of whether any of the five actually restores productivity, no comparison condition, and no measurement of the "user agency," "situational awareness," or "flexibility" they say the approaches should increase. What follows at that strength is a research agenda translating general automation-design principles into GenAI-specific proposals, not a validated remedy.