Interaction, Process, Infrastructure: A Unified Framework for Human-Agent Collaboration

TL;DR

This paper argues that current AI tools fail to sustain long-term, adaptive collaboration because they lack an explicit, editable representation of collaborative activity. It introduces Structural Adaptation as a core concept and a five-module Process Model (Problem Space, Workflow, Operations, Environment, Reflection) to support dynamic reconfiguration of work. A three-layer Interaction–Process–Infrastructure framework is proposed to centralize process representations and enable stable, cross-tool execution. The authors also present analytic lenses and discuss design challenges, opportunities, and tradeoffs for building process-first, open-ended human–agent systems with longitudinal co-evolution potential.

Abstract

While AI tools are increasingly prevalent in knowledge work, they remain fragmented, lacking the architectural foundation for sustained, adaptive collaboration. We argue this limitation stems from their inability to represent and manage the structure of collaborative work. To bridge this gap, we propose a layered conceptual framework for human-agent systems that integrates Interaction, Process, and Infrastructure. Crucially, our framework elevates Process to a first-class concern, an explicit, inspectable structural representation of activities. The central theoretical construct is Structural Adaptation, enabling the process to dynamically reorganize itself in response to evolving goals. We introduce a five-module Process Model as the representational basis for this adaptation. This model offers a unified theoretical grounding, reimagining human-agent collaboration as a coherent system for complex real-world work.

Figures (4)

• Figure 1: Collaborative activity unfolds through iterative cycles in which participants move between the problem space and the action space. Exploration generates insight that reshapes the problem formulation, while execution produces results that inform subsequent steps. This evolving structure motivates the need for an explicit process representation.
• Figure 2: The process as an explicit representation of collaborative activity. It consists of five structural modules: Problem Space, Workflow, Operations, Environment, and Reflection. They capture the evolving organization, context, and methods of the activity. This representational substrate enables structural adaptation across reframing, reorganization, and reinterpretation.
• Figure 3: Three-layer conceptual framework for process-first collaboration. The Interaction Layer provides user-facing projections of the activity; the Process Layer maintains the explicit, editable structure of the activity; and the Infrastructure Layer supplies the models, tools, memory, and resources that support execution and continuity.
• Figure 4: Representative forms of systems in today's AI landscape: chatbots, copilots, and agents. Each encodes a distinct model of collaboration but remains isolated, lacking connective logic across tasks and interfaces.