![[2412.10999] Cocoa: Co-Planning and Co-Execution with AI Agents](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2412.10999] Cocoa: Co-Planning and Co-Execution with AI Agents
Summary
The paper presents Cocoa, a system designed to enhance human-agent collaboration in AI tasks by allowing flexible co-planning and co-execution, enabling users to delegate tasks and adjust plans dynamically.
Why It Matters
As AI agents become integral to complex tasks, effective collaboration between humans and AI is crucial. Cocoa addresses the limitations of traditional workflows by promoting a more interactive and adaptable approach to task management, which can significantly enhance productivity and user experience.
Key Takeaways
- Cocoa enables flexible delegation of tasks between humans and AI agents.
- The system allows for interleaving planning and execution, enhancing adaptability.
- Lab and field studies demonstrate Cocoa's effectiveness in real-world applications.
- Cocoa's design is inspired by computational notebooks, facilitating complex research tasks.
- The approach promotes deeper collaboration and steerability without sacrificing usability.
Computer Science > Human-Computer Interaction arXiv:2412.10999 (cs) [Submitted on 14 Dec 2024 (v1), last revised 18 Feb 2026 (this version, v4)] Title:Cocoa: Co-Planning and Co-Execution with AI Agents Authors:K. J. Kevin Feng, Kevin Pu, Matt Latzke, Tal August, Pao Siangliulue, Jonathan Bragg, Daniel S. Weld, Amy X. Zhang, Joseph Chee Chang View a PDF of the paper titled Cocoa: Co-Planning and Co-Execution with AI Agents, by K. J. Kevin Feng and 8 other authors View PDF HTML (experimental) Abstract:As AI agents take on increasingly long-running tasks involving sophisticated planning and execution, there is a corresponding need for novel interaction designs that enable deeper human-agent collaboration. However, most prior works leverage human interaction to fix "autonomous" workflows that have yet to become fully autonomous or rigidly treat planning and execution as separate stages. Based on a formative study with 9 researchers using AI to support their work, we propose a design that affords greater flexibility in collaboration, so that users can 1) delegate agency to the user or agent via a collaborative plan where individual steps can be assigned; and 2) interleave planning and execution so that plans can adjust after partial execution. We introduce Cocoa, a system that takes design inspiration from computational notebooks to support complex research tasks. A lab study (n=16) found that Cocoa enabled steerability without sacrificing ease-of-use, and a week-long field dep...