A comparative study of 4,323 governance records from two rival standards for how AI agents find and trust each other, Ethereum's permissionless ERC 8004 trust protocol and Google's corporate led A2A (agent to agent) protocol, finds comparable participation inequality in both,…
When researchers pulled 4,323 governance records from two rival designs for how AI agents establish trust and share authority, one run on Ethereum's open blockchain and the other engineered by Google, they expected the open system to look more democratic. It did not. Both regimes showed similar participation inequality and community fragmentation. But the open, permissionless version produced a denser, more unified conversation among participants, a structural difference in discourse alignment that the corporate version did not replicate.
That finding, from a new comparative study on arXiv titled "Agentic Analysis for Agentic Infrastructure", is the actual news for anyone designing, auditing, or deciding between agent interoperability standards.
The two standards under examination sit at the center of a fast-emerging fight over who gets to define how autonomous AI software finds, trusts, and pays other autonomous AI software. ERC-8004, "Trustless Agents," is a draft Ethereum Improvement Proposal, a permissionless, on-chain standard for agent identity, reputation, and validation, with open reference contracts and a mainnet footprint reported by Forbes in February 2026. Google A2A, by contrast, is a corporate-led agent-to-agent interoperability protocol, engineered inside one of the largest AI vendors rather than by a token-holder community. The two represent the most visible rival designs for governing the same problem.
The paper's pipeline, consisting of LLM-assisted coding of governance discourse, neural topic modeling to surface themes, and multi-layer network analysis to map who talks to whom, is the method, not the story. Three findings from applying it to matched governance records across both regimes are what actually matter.
First, participation inequality is comparable: a small core of contributors drives most governance activity in both regimes, and that concentration is not meaningfully smaller in the open setting. Second, community fragmentation is comparable: the participation networks in both regimes break into distinct subgroups rather than a single coherent community. Third, and most consequentially, discourse alignment diverges: the topics discussed in the permissionless setting cluster more tightly together, while the corporate-led setting shows wider thematic spread across its subgroups.
The implication is not that decentralization wins. The paper's authors are explicit that openness does not, on its own, deliver broader participation or a less fragmented community. What it appears to deliver is a different kind of coordination: a denser convergence on what is being discussed, even when the same small set of participants is doing the discussing. For standards designers and auditors, that is a structural lever, not a slogan. If a use case depends on many contributors converging on shared vocabulary and shared priorities, say a safety-critical agent economy, the open track currently produces more of that. If a use case depends on a coherent corporate roadmap and predictable iteration cycles, the corporate track has its own logic.
The caveats matter. The dataset is a snapshot of two regimes at a specific point in their history; ERC-8004 was a draft EIP for most of that window, and A2A is a young standard. The method relies on LLM-assisted coding, which inherits whatever biases the underlying models bring to discourse interpretation. Replication materials, including code and data, were released alongside the paper, which makes the empirical claims testable, but the broader question of how these governance forms evolve under load remains open.
What to watch next: ERC-8004's mainnet footprint is recent enough that participation patterns could shift quickly if adoption grows, and A2A is still early enough that a single corporate revision could reshape its governance discourse. The cleanest test of the paper's claim will be whether the alignment gap persists when the open track attracts a larger and more diverse contributor base, or whether, as the corporate track iterates, it closes the discourse gap without narrowing the fragmentation gap.