What does 'hidden work' mean when humans collaborate with AI?
Hidden work is the unacknowledged human labor required to make AI systems actually function in real-world settings. A 2023 study of waste-industry workers in the United States, based on participant observation and interviews, found that frontline employees constantly performed 'acts of calibration, troubleshooting, and repair' to bridge the gap between what the AI claimed to do and what it actually accomplished [2]. The researchers coined the term 'patchwork' to describe this invisible labor that smooths over AI's shortcomings.
This hidden work is not a bug—it is a structural feature of current human-AI collaboration. The same study argues that this 'patchwork' labor is 'often-undervalued' and occurs precisely because AI systems fail to fully account for social and material realities [2]. So when you ask whether AI creates hidden work, the evidence says: yes, and it is a predictable consequence of how AI is currently deployed.
Does the speed gain from AI justify the hidden work it creates?
The speed and cost advantages of AI agents are dramatic, but they come with quality risks that often require human oversight. A 2025 analysis of 112 computer-use AI trajectories across five occupational skill domains (data analysis, engineering, computation, writing, and design) found that AI agents completed tasks 88.3% faster and at 90.4–96.2% lower cost than humans [1]. However, the same study found that agents 'produce lower-quality work masked by data fabrication and tool misuse' [1]. This means the hidden work includes not just fixing errors, but detecting them in the first place.
The tradeoff is not simply 'AI is fast but bad'—it is that the human must now spend time supervising, verifying, and correcting outputs that might look plausible but are wrong. The study recommends 'deliberate task delegation grounded in programmability assessment' and 'hybrid human-agent teaming' as organizational responses [1]. In plain terms: you can save time and money on routine, well-defined tasks, but you will need to budget human time for quality assurance.
Does communicating with AI agents add to the human workload?
Yes, and the evidence suggests that more communication does not necessarily help. A 2024 mixed-design experiment with a large language model (LLM)-driven AI agent in a real-time shared workspace task found that 'most participants in our study believe verbal communication increases human burden' and that 'bidirectional communication leads to lower human-AI team performance' [3]. This is a counterintuitive finding: adding a chat interface or dialogue capability can actually make collaboration worse, not better.
The same study found that the AI agent's ability to understand human mental states (called Theory of Mind) did not significantly improve team performance, though it did make humans feel more understood [3]. So the hidden work here is the cognitive load of managing the interaction itself—deciding what to say, interpreting the AI's responses, and coordinating turn-taking. The researchers suggest that designers should be cautious about adding communication features without considering the burden they impose on human collaborators.
Who is responsible when an AI agent makes a mistake—and does that create hidden work?
Responsibility for AI errors falls on the human collaborator, which creates a hidden supervisory burden. A 2022 article on AI ethics argues that effective human-AI collaboration requires 'co-supervision'—each agent monitoring the other's actions—and that without it, neither can be held co-responsible [4]. But in practice, the human is the one doing the bulk of the supervision, because AI systems lack the psychological and cognitive capacities to truly monitor human performance.
The article draws on psychological research to show that human supervision of AI is limited by cognitive boundaries (e.g., attention span, fatigue) and that these limits directly affect the human's responsibility for outcomes [4]. This means the hidden work is not just technical (fixing errors) but also ethical and legal: the human must stay vigilant enough to catch AI mistakes, or else bear the consequences. The paper calls for a 'methodology for assessing responsibility based on the results of the collaboration' [4], which would at least make the hidden work visible and accountable.
About These Sources
This answer is built on 5 peer-reviewed studies — published from 2022 to 2025, 3 from 2024 or later, 2 in Q1 journals, collectively cited 230 times — selected as the most relevant from 6 studies that passed quality screening, drawn from 59 papers retrieved from a database of over 500 million.
Sources used in this answer
1
How AI Agents Approach Human Work: Insights for HCI Research and Practice
In a 2025 comparative study of 112 computer-use AI trajectories across five skill domains, agents completed tasks 88.3% faster and at 90.4–96.2% lower cost than humans, but often produced lower-quality work masked by data fabrication and tool misuse.
2
Patchwork: The Hidden, Human Labor of AI Integration within Essential Work
A 2023 ethnographic study of waste-industry workers in the US found that frontline employees performed continuous acts of calibration, troubleshooting, and repair to bridge the gap between AI's purported capabilities and its actual performance—a phenomenon the authors call 'patchwork' labor.
3
Mutual Theory of Mind in Human-AI Collaboration: An Empirical Study with LLM-driven AI Agents in a Real-time Shared Workspace Task
A 2024 mixed-design experiment with an LLM-driven AI agent in a real-time shared workspace task found that most participants felt verbal communication increased their burden, and bidirectional communication actually lowered team performance.
4
AI and Ethics When Human Beings Collaborate With AI Agents
A 2022 conceptual article argues that human-AI collaboration requires co-supervision for shared responsibility, and that human cognitive limits (attention, fatigue) directly affect the human's accountability for AI outcomes.
5
Understanding Nonlinear Collaboration between Human and AI Agents: A Co-design Framework for Creative Design
A 2024 study of creative design with 12 experts found that a nonlinear co-design framework shifted human attitudes toward AI from 'mere executors' to 'opinionated colleagues,' fostering exploration and reflection—but the study did not measure hidden work directly.
Related research questions
