Researchers want to make agent identity a formal design discipline, not an afterthought.
A paper published on arXiv introduces the Agentic Social Affordance Framework, or ASAF, a theoretical model built on the idea that the social identity of an AI agent — its name, role framing, and perceived personality — functions as a collaboration interface. The framework proposes three mechanisms through which identity does this work: Identity Signaling, Behavioral Priming, and Collaborative Governance. It also introduces a spectrum for measuring how faithfully an agent's signals communicate its identity, and accounts for individual differences in how users respond — specifically whether someone tends to anthropomorphize AI or treat it as a pure tool.
The practical implication is that identity design is not a UX flourish bolted on after the engineering is done. In multi-agent systems where humans are still in the loop, the way each agent presents itself shapes what operators notice, what they question, and how much autonomy they grant. The authors draw a pointed contrast with classical multi-agent systems research, which used roles and norms to constrain autonomous agents; ASAF flips the direction and applies that same vocabulary to structure human cognition instead.
For teams building agentic pipelines — which are multiplying fast across enterprise software — this suggests that the gap between a well-orchestrated system and an effective one may have less to do with the models underneath and more to do with how the agents are introduced to the people working alongside them. The framework is still theoretical and the authors call for empirical validation, so for now treat it as a useful lens, not a tested playbook.