Sociological AI Architecture
Stagecraft operationalizes 90 years of sociological theory — Mead, Goffman, symbolic interactionism — into a production-ready coordination layer for multi-agent AI systems. Built on existing transformer infrastructure. 10–20% compute overhead. 60%+ improvement on compositional reasoning tasks.
The Theory
The problem
Compositional generalization, context-switching, and common sense — all stalled. Not because of insufficient compute or data. Because the architecture is wrong. They are building individual agents.
The insight
Human intelligence scaled with social group size, not habitat difficulty. Dunbar's social brain hypothesis. Perspective-taking and role-adoption are not soft skills — they are the core machinery of intelligence.
The source
George Herbert Mead (1934) and Erving Goffman (1959) identified the mechanisms. These have never been operationalized into AI architecture. Stagecraft does that.
"The self is not something that exists first in寂静 and then gets put into a situation — it is something that exists only in the context of social experience."— George Herbert Mead
Three-Layer Architecture
Multiple agent perspectives generate parallel interpretations of context. The ensemble, not the individual agent, holds the full picture. Inspired by Mead's concept of taking the attitude of the generalized other.
The coordinator assigns roles dynamically based on task requirements and agent specialization. Role-adoption is computed, not prompted. Based on Goffman's role theory and dramaturgical analysis.
Output adapts to audience context — different framing for a compliance officer than a developer, different precision for a CFO than a claims adjuster. Goffman's performative self, computational.
Where it matters
370,000 submissions a month. Context-switching between risk assessment, compliance review, and broker communication is where current orchestration fails. Stagecraft handles it.
Multi-step reasoning across regulatory contexts, different stakeholder audiences, and evolving market conditions. The audience-adaptive layer alone is worth the architecture.
Role-adaptive performance across paralegal research, attorney drafting, and partner review. Different outputs for the same underlying reasoning — no additional compute, just better routing.
ERP and CRM integration with cross-functional agent coordination. Orchestration across front, mid, and back office without adding headcount — the AIG playbook, now with social intelligence.
Stagecraft is not another framework. It is a fundamentally different architecture — grounded in the same theory that explains why human intelligence scaled when it did. The multi-agent orchestration market is growing at 48.5% CAGR. The winners will not be the teams with better graphs. They will be the teams who understand why coordination actually works.
Contact
Interested in Stagecraft for your organization? Reach out directly.