If an AI teammate's job keeps changing mid-project, should it ever be the one reassigning tasks to you?
How should AI roles adapt when team tasks become dynamic and uncertain?
This explores how the jobs we give AI inside a team (assistant, worker, coordinator, decision-maker) should shift when the work keeps changing and nobody knows the right plan in advance, and what the corpus says about building teams that can reorganize themselves as they go.
This explores how AI's role in a team should change when the task won't hold still: when plans shift partway through, the right expertise isn't obvious up front, and nobody can say ahead of time who should do what. The short answer from the corpus is that fixed roles are the wrong design. What works is treating roles as things you can reassign mid-task, sometimes in both directions between humans and AI.
The most surprising thread is that delegation doesn't have to flow only from humans to AI. Can AI systems delegate tasks back to humans? argues that agentic systems can hand tasks and responsibilities back to people, so the AI sometimes acts as the assigner rather than the assignee. That matches what executives report: in one large survey, most saw agentic AI as a coworker rather than a tool, and the tension comes from the fact that it is both at once Does agentic AI fit the tool-or-coworker framework?. In uncertain work, the useful question changes from "what is the AI's role?" to "how does the role get renegotiated as things change?" How should humans and AI systems divide decision-making tasks? treats this as an organization-design problem: how you split tasks and how each side learns from feedback should depend on what kind of task it is.
Several systems make that renegotiation concrete. In Cocoa, users can assign each plan step to themselves or to the agent and revise the plan while it runs. In a small lab study, this made the agent easier to steer than chat without making it harder to use Can users steer AI agents better by revising plans mid-execution?. Magentic-UI starts from a frank admission: there is no ground truth for exactly when an agent should stop and ask a human. So instead of solving that timing problem, it spreads checkpoints across six mechanisms, including co-planning, action guards and verification When should human-agent systems ask for human help?. The lesson carries over well: when the right moment to hand off is unknowable, build many small handoff points instead of one perfect rule.
On the AI-to-AI side, the same idea shows up as teams that rebuild themselves. DyLAN scores each agent's contribution while the task runs and switches off the ones that aren't helping, so the team's makeup adapts without task-specific tuning Can multi-agent teams automatically remove their weakest members?. IMACS separates three things that most frameworks bundle together: who plays which role, how agents coordinate, and how their outputs get combined. That lets you swap one without rewriting the others Can multi-agent teams swap roles, coordination, and algorithms independently?. There's also a reason not to just make one agent smarter: tasks that need varied expertise, parallel work and independent checking hit an organizational ceiling that extra capability or context doesn't fix Do single agents always hit organizational limits?. One warning about labels: assigning an agent "accountability" only changes outcomes if the workflow actually routes the work through that agent, and the best placement varied by model family Does accountability placement actually shape multi-agent outcomes?. A role exists only if the control flow enforces it.
Two cautions keep this grounded. Today's agents still struggle in realistic workplaces: the best ones finished only about 30% of simulated office tasks on their own, and social interaction was one of the main failure points Why do AI agents fail at workplace social interaction?. They also tend to wait for instructions instead of speaking up, because training that rewards each immediate reply removes initiative. Initiative can be trained back in, but it has to be balanced against being intrusive Why do AI agents fail to take initiative?. So as AI roles become more fluid, oversight should come first: Must AI monitoring precede incentive alignment as agents evolve? argues that monitoring has to be in place before you can trust incentive-style alignment, because AI decisions get harder to read as autonomy grows. The unexpected takeaway is that adaptive teams depend less on smarter agents and more on good plumbing: reassignable steps, frequent check-ins, and routing that gives roles real effect.
Sources 12 notes
Holldack, Banh and Strobel argue that agentic information systems capable of independent decision-making can transfer rights and responsibilities to humans, reversing the traditional one-directional delegation model. This requires moving beyond agency-primacy frameworks to delegation theory that accounts for bidirectional task assignment.
A survey of 2,102 executives across 116 countries found 76% view agentic AI as a coworker rather than a tool. This duality creates operational tensions because agentic systems simultaneously substitute and complement labor, requiring both HR and asset management approaches that traditional either-or frameworks cannot accommodate.
Puranam frames human-AI pairs as multi-agent organizations requiring design choices about task division and learning configuration. He distinguishes three task types and shows learning mechanisms produce different performance outcomes based on feedback structure and decision architecture.
Cocoa's system, which lets users assign plan steps to themselves or an agent and revise plans mid-execution, afforded greater steerability than chat-baseline interaction in a lab study (n=16) while maintaining ease of use, with field users valuing explicit work delegation.
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.
Show all 12 sources
DyLAN's three-step importance scoring mechanism (propagation, aggregation, selection) quantifies individual agent contributions and automatically removes uninformative agents during inference, optimizing team composition without task-specific tuning.
IMACS demonstrates that team structure (Belbin roles), coordination mechanism (Mintzberg framework), and fusion algorithm can be architecturally separated into independent components. This decoupling lets teams and protocols swap without rewriting each other, unlike existing frameworks that fuse these concerns together.
Research shows that real-world tasks requiring heterogeneous expertise, parallel execution, and independent verification exceed what any single agent loop can organize. Graph-based system abstractions are needed to distribute intelligence across specialized agents.
Ablation studies show RACI accountability tags change outcomes only when the collaboration protocol's control flow passes the deliverable through the accountable agent. The best placement varies by model family, requiring per-model validation rather than universal design.
TheAgentCompany benchmark shows leading agents achieve 30% task completion in a simulated workplace. Social interaction, professional UI navigation, and domain-specific knowledge are the three primary failure modes, with multi-turn task performance consistently dropping to 35% across enterprise settings.
Research shows next-turn reward optimization structurally removes initiative from models, but proactive behaviors like critical thinking and clarification-seeking are trainable (0.15% to 73.98% with RL). The core challenge is balancing proactivity with civility to avoid intrusion.
Humberd and Latham argue that as AI becomes more autonomous, monitoring mechanisms must be established first as prerequisites to incentive alignment, because defining desirable AI behaviors is harder than for humans and AI decision-making becomes increasingly incomprehensible over time.
Papers this line draws on 8
The research behind the notes this line reads — ranked by how closely each paper relates.
- HumanAgencyBench: Scalable Evaluation of Human Agency Support in AI Assistants
- Towards a Science of Scaling Agent Systems
- Worse Together: How Performance Breaks Down in Multi-User Multi-Agent Teams
- Unraveling Human-AI Teaming: A Review and Outlook
- Agentic AI and Human-in-the-Loop Interventions: Field Experimental Evidence from Alibaba's Customer Service Operations
- When AI Becomes an Agent of the Firm: Examining the Evolution of AI in Organizations Through an Agency Theory Lens
- Toward an Organizational Science of Multi-Agent LLM Systems: Decoupling Who, How, and Which Algorithm
- Cocoa: Co-Planning and Co-Execution with AI Agents