Philosophers from Socrates to Bertrand Russell have underscored that genuine agreement arises not from superficial accord but from reasoned dialogue that harmonizes diverse viewpoints. Jürgen Habermas’s theory of communicative action refines this principle into a vision of discourse aimed at consensus through rational argument. Recently, a paper in Science by Michael Henry Tessler et al. (2024) (“AI can help humans find common ground in democratic deliberation”) echoes this idea by describing a “Habermas Machine”—an AI mediator capable of synthesizing individual opinions and critiques to foster mutual understanding. While their study focuses on social and political issues, the underlying concepts extend readily to organizational contexts and knowledge management.

In our own effort to realize a Habermas-inspired mediator, we employ an architecture that leverages BigQuery as a data warehouse built on a Data Vault schema, managed and orchestrated with dbt (Data Build Tool). The system ingests communications from platforms such as Slack and Gmail, breaking each message into paragraph-level segments for individual vector embeddings. These embeddings are then stored in BigQuery, forming a semantic layer that augments traditional relational queries with more nuanced linguistic searches. In the below diagram, you can see how messages flow from raw capture to an enriched, queryable knowledge graph.

This structural framework, however, only solves part of the puzzle. We then introduce LangGraph agents, enhanced by tooling such as LangSmith, to marry textual and structural data. These agents can retrieve messages based not only on metadata (author, timestamp) but also on thematic or conceptual overlap, enabling them to detect undercurrents of agreement or contradiction in vast message sets. In a second diagram, below, you can see how agent-mediated queries integrate semantic vectors, user roles, and conversation timelines to pinpoint salient insights or latent conflicts that humans might overlook.

The philosophical impetus behind this design lies in extending what Habermas posits for face-to-face discourse—an “ideal speech situation”—to distributed, digitally mediated communication. Like the “Habermas Machine” described by Tessler et al., our system provides prompts and syntheses that help participants recognize areas of accord and legitimize points of dissent, rather than imposing a solution from on high. A final diagram, below, depicts a feedback loop, where humans validate or refute AI-suggested statements, gradually converging on well-supported, collectively endorsed conclusions.

Ultimately, these tools do not replace human judgment; they aspire to enhance it. By combining robust data engineering on BigQuery with sophisticated natural-language reasoning via LangGraph agents, we strive to ground the ideal of rational consensus in a practical, scalable system. Inspired by recent research and Habermasian philosophy, we envision AI as a diplomatic catalyst—one that quietly structures and clarifies discourse, guiding us toward common ground without diluting the richness of individual perspectives.