How do communication modalities shape human-agent collaboration patterns?

Most human-agent collaboration research measures whether agents complete tasks. This platform takes a different approach: adapting classic human-human collaboration experiments (the Shape Factory paradigm from CSCW) to systematically manipulate the conditions of collaboration, not just its outcomes.

The platform enables researchers to independently manipulate three theory-grounded interaction controls:

1. Communication modality — How humans and agents exchange information (text, voice, structured channels). Varying this produces distinct patterns in negotiation frequency and team performance.
2. Awareness dashboards — What information about agent state, progress, and reasoning is visible to the human collaborator. Different awareness levels produce different trust patterns.
3. Social framing — How the agent is presented (tool, teammate, partner). This affects how humans apply social scripts from When should human-agent systems ask for human help?.

Key finding from validation studies (16 participants, crossed between-subjects design): varying communication modality produced significant differences in perceived trust and workspace awareness, aligning with established CSCW findings for human-human collaboration. This suggests human-agent collaboration may follow similar structural patterns to human-human collaboration when interaction controls are properly configured.

The platform architecture has four components: researcher interface for parameter manipulation, modularized participant interface, standardized agent context protocol, and experiment controller with logging. This modularity allows reimplementation of classic experimental paradigms — strategic collaboration (DayTrader), collaborative decision-making (Essay Ranking), and task-solving (Passcode).

This insight is methodologically important: it shifts the research question from "can agents collaborate?" to "under what conditions does collaboration work, and which controls matter most?" The finding that communication modality affects negotiation frequency connects to Why do AI agents misalign with what users actually want? — the UserBench finding may be partly an artifact of text-only, chatbot-style interaction rather than an intrinsic agent limitation.