"Nobody Did This": Contribution, Originality, and Accountability in Agent-Mediated Collaboration

Abstract

Collaborative knowledge work is changing in ways that go deeper than disclosure or transparency. LLM agents are now embedded in how teams research, design, write, and decide: mediating between team members, synthesizing inputs, reformulating ideas, and drafting shared outputs. They do not only facilitate collaboration; they operate within the collaborative workflow at the moment when contributions are being formed. In doing so, they risk undermining the social conditions under which contribution can be witnessed, attributed, and held accountable.

This workshop brings together researchers and practitioners to confront what we call contribution dissolution: the blurring of attribution, originality, and accountability in agent-mediated collaborative work. We argue that this dissolution begins before collaboration itself, in the individual worker's own uncertainty about what, in their agent-mediated work, is genuinely theirs, and propagates through collaborative relationships, collapsing the reliability that makes productive intellectual exchange possible.

Through provocative position statements, hands-on mapping exercises, and the workshop activity, participants will surface how framing accountability in collaborative work as a documentation problem (e.g., AI use statements, watermarking, provenance logs) overlooks the conditions under which accountability is produced. Our goal is to produce a shared research agenda and the foundations of an infrastructural response to contribution dissolution in collaborative knowledge work.

Introduction

Collaborative knowledge work rests on an implicit social contract: people who work together maintain a shared, informal record of who contributed what. This record lives in memory, in conversation traces, in the history of drafts and revisions. Drawing on Suchman's account of accountable action as situationally embedded, we call this witnessed contribution: a contribution that can be attributed, contested, and accounted for because it emerged within a shared social situation that made it intelligible to others.

Witnessed contribution is not the same as documented provenance. Provenance records that something was produced and how it was produced. What makes a contribution attributable and accountable is that it was produced under conditions where others could recognize it as an act of judgment, respond to it, and hold the contributor(s) answerable. A provenance log cannot show whether a framing was the person's or the agent's, or whether the person could defend it under challenge. These require the kind of situated social interaction that agent mediation risks undermining.

LLM agents shift this at the individual level first. Rather than using agents to externalize already-formed ideas, people co-think with them, with ideas emerging through interaction. They draft, refine, and develop lines of reasoning in dialogue with these agents. As a result, the boundary between what the individual contributed and what the agent introduced becomes difficult to reconstruct after the fact. It was never established during the process itself. A worker may have procedural ownership (they prompted, reviewed, and accepted) without having substantive authorship that could survive scrutiny. The individual worker arrives at the collaborative table already unable to reliably say what is theirs.

At the collaborative level, agents intervene at the moment contributions are formed and shared. When a team member uses an agent to reformulate an idea before sharing it, when an agent reconciles two collaborators' positions into a synthesized output, when an agent drafts a deliverable from multiple inputs, the contribution that reaches collaborators is not what the person said or wrote but what the agent produced from what they said or wrote. Social creativity depends on stakeholders externalizing genuinely distinct perspectives; it is the friction between different framings that generates new understanding. Agents are optimized to follow instructions, not to resist them. The result is convergence: outputs that reflect dominant assumptions rather than the situated insight of the person prompting.

When a colleague shares agent-mediated work, the recipient cannot read it as a reliable signal of that colleague's thinking. They do not know whether the contributor can defend it, whether their response to pushback represents a real intellectual position or a new prompt, or whether the voice in the document belongs to the person or the model. The contributor themselves may not know. We call this contribution dissolution: not a failure of disclosure but a condition in which the social basis for attribution, originality, and accountability has been blurred.

Common responses (disclosure-based solutions, AI use statements, watermarking, provenance tracking) frame contribution dissolution as a problem of attribution and documentation: if we could better track what the agent did, we could restore accountability. This framing is productive in cases where human and agent contributions remain separable. However, when a team member uses an agent to "clean up" collaborative notes before sharing them, the cleanup may restructure the argument in ways that redistribute credit for key ideas. Watermarks may not persist through ordinary collaborative refinement. Detection tools fail in tightly coupled human–agent work where outputs converge. Disclosure cannot surface what was never witnessed.

Workshop Themes

The workshop is organized around three threads:

• Contribution Dissolution. How do agents restructure the process through which contributions are formed, and what is lost when they intervene inside the collaborative loop? Both at the individual level (the worker's uncertainty about what is genuinely theirs) and the collaborative level (collapse of productive asymmetry between contributors when agent refinement makes distinct perspectives indistinguishable).
• The Documentation Trap. How do documentation-based responses (AI use statements, detection tools, watermarking, provenance tracking) fail, and what does this failure reveal? Our central argument is that these solutions approach contribution dissolution as a problem of missing information. They assume that the evidence needed for attribution and accountability exists and merely needs to be surfaced.
• Accountability Infrastructures. How should we think about accountability relations given agent-mediated work? What might the composition of appropriate forums look like to judge agent-mediated work? If documentation can no longer be reliable sources of attribution due to blurred human–agent boundaries, what would accountability infrastructures require?

Workshop Activities

Main workshop schedule

The Documentation Trap activity

Participants receive a package of materials representing a completed collaborative project: a final report, a set of intermediate drafts, and a provenance log recording all agent interactions throughout. Groups are asked to answer four questions from the materials alone:

• Who had the original idea for the central framing?
• Whose position won when two collaborators disagreed?
• Which contributions survived agent reformulation, and which were lost?
• Who should be held accountable if the report's central claim turns out to be wrong?

Groups will find they cannot answer these questions from the provenance record. The debrief surfaces why: this is not a failure of the provenance log. It is a limitation of what the log was asked to record. The activity feeds directly into the afternoon session on accountability infrastructure.

Before the workshop

We aim for ~25–30 participants, balancing academic and practitioner perspectives. A small Program Committee will review each submission with at least two PC members, focusing on concreteness of the case, engagement with workshop themes, and potential for discussion. We will prioritize submissions that present empirical encounters with contribution dissolution from practice, research, or professional experience rather than purely theoretical work.

Accepted participants will be invited to a dedicated Slack channel where they will share brief introductions and a short summary of their position provocation, allowing them to familiarize themselves with each other's cases.

Post-Workshop Plans & Dissemination

All accepted materials (position provocations, slides, and artifacts generated during the workshop) will be made available on the workshop website under a CC BY license. The primary output will be a collaboratively developed Accountability Infrastructure Casebook: a design provocation, grounded in the concrete cases participants bring, outlining what collaborative systems and institutional norms would need to be preserved for witnessed contribution to remain possible.

To reach a wider audience, we will archive accepted position provocations and workshop artifacts with DOIs on Zenodo under an open license. In the long term, we will synthesize the findings into a collaboratively authored paper for CSCW or CHI and disseminate a summary in ACM Interactions. Participants will be invited to continue through a post-workshop working group focused on the governance implications of contribution dissolution across professional and institutional settings.

Accessibility

Accessibility will be integrated throughout the workshop. Authors will be required to follow the SIGCHI Accessible Submission Guide. We will review submissions for accessibility compliance, collect information about participant accessibility needs in advance, and coordinate with the CSCW Accessibility Chairs during the event. All workshop materials will include alt text and other accessible features.