Can the same AI be a true teammate on one task and just a tool on the next — does that depend on the task, not the AI?
Can a machine be cooperative in one routine but merely assisting in another?
This explores whether a machine's level of agency belongs to the machine itself or to each task it takes part in, so that the same system could be a true partner in one workflow and a passive helper in another.
This explores whether 'cooperative' versus 'assisting' describes a machine or describes a particular routine it takes part in. The corpus points toward the second answer: agency is a property of the arrangement, not a fixed trait of the technology. One framework places machine agency on five levels: passive, semi-active, reactive, proactive and cooperative Does machine agency exist on a spectrum rather than binary?. Nothing in that spectrum says a system must stay on one level. The same AI that only answers when asked in one workflow can plan alongside you in another, depending on what that routine lets it do.
Organization research makes this concrete. Murray, Rhymer and Sirmon sort human-technology routines by two separate questions: who sets the rules (the protocol), and who chooses the next action How do humans and technology share control over organizational routines?. Because these are two separate controls, one tool can sit in different spots in different routines. It might follow human-written rules while picking actions in one process, and simply carry out human choices in another. The framework also shows that this placement changes how well each routine adapts to change and feedback. So the difference between 'cooperative' and 'assisting' affects outcomes, not just labels.
Agent design takes the idea down to the level of moments within a single task. Magentic-UI doesn't fix one autonomy level. It switches between co-planning, co-tasking, action guards and verification steps When should human-agent systems ask for human help?. One session can move from partner (planning together) to supervised assistant (waiting for approval before a risky action) and back. The case for keeping humans in the loop until agents are reliable on open-ended work rests on the same logic: let the system cooperate where tasks are structured and grounded, and keep it assisting where judgment is needed Should AI systems stay collaborative rather than fully autonomous?.
The surprising part is that the routine may matter more than the machine's ability. A meta-analysis of more than 100 studies found that human-AI pairs usually do worse than the better of the two working alone on decision tasks, but gain on content-creation tasks When do human-AI teams actually outperform either alone?. The same pairing helps in one kind of work and hurts in another. Even observers struggle to tell the modes apart. Process traces clearly reveal wholesale delegation, but ordinary collaborative help looks almost the same as working alone Can process data distinguish AI delegation from ordinary collaboration?.
The takeaway is that 'what level of agency should this AI have?' is the wrong unit of analysis. A better question is 'what role should it play in this routine?' Economics adds a twist: when one step of a job is automated, the human steps that remain can become more valuable, not less Does automating some tasks actually increase the value of remaining work?. So choosing where a machine cooperates and where it only assists also changes what the human's share of the work is worth.
Sources 7 notes
Research shows machine agency ranges across five levels—passive, semi-active, reactive, proactive, and cooperative—rather than existing as a binary choice. Users experience and judge these interactions through a 'machine heuristic' mental shortcut, and the mismatch between what AI can do and what workers want reveals deployment opportunities.
Murray, Rhymer and Sirmon's 2021 framework divides shared agency by whether humans or technology set protocols and select actions. This split determines how responsive a routine is to change and feedback.
Magentic-UI identifies co-planning, co-tasking, action guards, verification, memory, and multitasking as mechanisms that work around the lack of ground truth for optimal deferral timing. Rather than solving the timing problem directly, these mechanisms distribute decision-making across multiple touchpoints.
Collaborative systems where humans remain in the loop outperform autonomous agents on hallucination correction, ambiguity resolution, and accountability. Evidence shows AI is reliable only on structured, retrieval-grounded tasks, not novel research or judgment.
A meta-analysis of over 100 studies finds human-AI combinations score lower than the best of human or AI alone (g = −0.23), except in content-creation tasks where gains emerge. Decision tasks show losses; the difference lies in task structure, not ability.
Show all 7 sources
Analysis of writing and programming corpora shows AI contributions arrive in concentrated bursts outside authors' baseline rhythms, creating a categorical signature for wholesale delegation while leaving collaborative assistance indistinguishable from minimally assisted work.
When job tasks multiply together rather than add (O-ring production), automating some tasks concentrates worker time on what remains, raising its quality and value. This means standard exposure indices underestimate worker resilience by assuming static task value.
Papers this line draws on 8
The research behind the notes this line reads — ranked by how closely each paper relates.
- When combinations of humans and AI are useful: A systematic review and meta-analysis
- Cocoa: Co-Planning and Co-Execution with AI Agents
- Levels of Autonomy for AI Agents
- The Work Behind Delegation: A Framework for Supervising AI Coding Agents
- Does AI Assistance Leave a Temporal Fingerprint? Detecting Overreliance in AI-Assisted Writing and Programming
- O-Ring Automation
- Humans and Technology: Forms of Conjoined Agency in Organizations
- Complementarity in Human-AI Collaboration: Concept, Sources, and Evidence