Does generative AI actually save workers time or intensify it?

UC Berkeley Haas doctoral student Xingqi Maggie Ye and Associate Professor Aruna Ranganathan ran an eight-month ethnography at a 200-person U.S. technology company, observing work in real time and conducting "more than 40 semi-structured interviews across functional groups." Their finding, reported in Harvard Business Review, is that "generative AI didn't free up time—it expanded what workers felt capable of, and willing, to take on." They write that "employees worked at a faster pace, took on a broader scope of tasks, and extended work into more hours of the day, often without being asked to do so."

The researchers describe three mechanisms of this intensification. First, scope crept: "people began taking on work that previously would have belonged to someone else or might not have been attempted at all," so "the scope of what counted as 'my job' widened." Second, stopping points dissolved: because AI "makes it easy to start and continue tasks," work "seeped into moments that used to function as pauses" — prompts sent at lunch, before meetings, in the evening. Third, threads multiplied: workers "run AI processes in the background while reviewing code, drafting documents, or attending meetings," some "running multiple AI agents simultaneously," so that "both the human and the machine were constantly in motion." Ye argues this produces a vicious cycle: "increased capability leads to increased output, which leads to higher expectations, which then pressures further expansion," with constant switching eventually impairing judgment and raising errors. Ye also notes a split between moment-to-moment experience (momentum, expanded capability) and retrospective experience (feeling "busier, more stretched, or less able to fully disconnect").

This is a different claim from Does AI really save time, or just change how we spend it?, which found total time-on-task roughly constant while its composition shifted toward prompting and evaluation. Ye and Ranganathan's ethnography finds the total expanding — more hours, more scope, more concurrent threads — because AI lowers the cost of starting one more task rather than merely changing what a fixed task consists of. The two findings are compatible (time within a task can reallocate while the number of tasks a worker takes on also grows) but neither implies the other, and the Haas study is the one that locates the extra time specifically in blurred boundaries and self-driven scope expansion rather than in the mechanics of using the tool. Ye's warning that an "AI practice" needs deliberate human grounding "so work doesn't become entirely solo and tool-mediated" echoes the concern in Does generative AI shift knowledge workers away from communication?, but frames the drift toward solo work as a risk to be managed through pacing and check-ins rather than as an observed trace-data pattern.

The excerpt is a single ethnography at one 200-person company, described by its own authors as "in-progress research" without a published, peer-reviewed paper cited, and the mechanisms are drawn from interview self-report and observation rather than measured hours or output logs — it does not quantify how much additional time or how many additional tasks resulted, nor does it establish the pattern holds outside this one organization or this period of generative-AI adoption. The implication the authors draw, that organizations risk mistaking self-driven intensification for a durable productivity gain, follows reasonably from what they observed but should be read as a hypothesis this study motivates rather than one it has tested at scale.