Send message to Author

Paper short abstract

This paper introduces the concept of adequacy regimes to describe the socio-technical arrangements through which actors (or engineering more specifically) decide when knowledge or system outputs are good enough to proceed with work.

Paper long abstract

Artificial intelligence systems are increasingly integrated into everyday work practices, yet their outputs remain uncertain, strange, just weird, or often difficult to interpret. As a result, workers must continuously determine when AI-generated results are sufficiently reliable to incorporate into their tasks. This paper introduces the concept of adequacy regimes to describe the socio-technical arrangements through which actors decide when knowledge or system outputs are good enough to proceed with work.

Drawing on ethnographic research with robotics engineers, software developers, and technical practitioners working with AI systems, the paper examines how workers negotiate the epistemic expectations introduced by contemporary AI tools. Rather than simply automating decision-making, AI systems often shift the burden of judgment onto workers, who must determine when model outputs are acceptable, when further verification is required, and when the system should be ignored or overridden. These judgments are rarely individual decisions; instead they are shaped by organizational constraints, professional norms, engineering practices, and the epistemic logics embedded in AI technologies themselves.

Adequacy regimes emerge at the intersection of these forces, structuring acceptable levels of uncertainty, error, and responsibility in AI-mediated work. Through practices such as iterative testing, prompt experimentation, cross-checking, and collaborative troubleshooting, workers transform uncertain model outputs into actionable knowledge.

By focusing on adequacy regimes, this paper contributes to STS discussions about the epistemology of AI and the future of work. It shows how AI does not simply introduce new tools into workplaces but reshapes how workers evaluate knowledge, distribute responsibility, and define when work is complete.