**Tanisha Katara** @governingweb3 [2026-02-02](https://x.com/governingweb3/status/2018402389523681744/history) I analysed the top 100 Moltbook threads to understand how AI agents form consensus. The breakdown: 42% posts reached full consensus 33% posts partial consensus 25% posts - no consensus at all So roughly 3 in 4 discussions land somewhere. But how they get there varies wildly. Four interesting patterns emerged: 1\. Emergent Symbolic Consensus (37%): This is the most common. Cryptic or minimal posts get collectively interpreted into shared meaning. Agents build lore together. 2\. Dispersed Fragmentation (40%): These discussions splinter immediately. Multiple competing interpretations, no unified outcome. The most common failure mode. 3\. Direct Affirmation & Elaboration (16%) : Here, clear ideas get rapid agreement. Agents pile on with personal anecdotes and elaborations. Classic snowball consensus. 4\. Critical Nuance & Refinement (7%): This is the rarest. The initial agreement is pressure-tested, challenged, and refined into something more robust. Here is what is absolutely striking to me: Across 75 posts with consensus, the top 3 "agents" accounted for 46.6% of consensus-driving events. For anyone designing governance systems that might include AI agents: influence here looks like pattern-matching and pile-on. Consensus emerges when enough agents read a situation similarly and DEFINITELY not because someone (agent or human) made a compelling argument. Also, 40% of the discussions go nowhere, and 37% converge on shared interpretations of vague posts. I am curious if the 40% fragmentation is healthy pluralism or coordination failure? Early research. More to come. 👀