AGORA

The researchers proposed a framework called AGORA (Agents Growing through Organization and Relational Arrangements), which studies how social conditions shape teams' evolutionary trajectories by independently varying team composition and interdependence structures (such as collaboration, conformity, and competition).

The AGORA framework unfolds along two independent dimensions: team composition (determining who is on the team, i.e., the agents' capability configuration and role assignment) and interdependence structure (defining the interaction relationships among members, such as egalitarian collaboration, hierarchical conformity, or competitive confrontation).

The AGORA framework was cross-validated under three evolutionary mechanisms, covering three types of tasks: web search, codebase understanding, and document question answering.

When social scaffolding is carefully designed, team performance can exceed the 'strong individuals evolving alone' baseline by as much as 25 percentage points.

In the AGORA study, authority tends to concentrate in the most capable member of the team, while other members correspondingly 'step back'; this distribution of power is not artificially designated but naturally emerges from interaction.

The 'strongest member' that grew up in a team environment performs beyond the level of the same model running alone, meaning that the team experience itself changed the agent's capabilities.

The team structure in the AGORA framework is not a static configuration, but continuously reshapes the interaction patterns among members as evolution proceeds.

The study currently covers three types of tasks: web search, codebase understanding, and document question answering; its applicability in more open and more complex real-world scenarios remains to be verified, and whether the optimal configuration of social scaffolding changes dynamically with task type and team size is also a question that subsequent research needs to answer.