At a Glance
Anchor cross-team AI communication to each person’s persistent ledger so agents can share verifiable context directly, cutting costly human relay and giving a clear audit trail for agent behavior.
ON THIS PAGE
What They Found
A ledger-backed, person-centered communication model lets short-lived agents hand off work without humans acting as the transport layer. This aligns with the A2A Protocol Pattern. An always-on lightweight gateway (an "embassy") stores inbound messages durably, so any agent can resume tasks from that shared record. This mirrors the Agent Service Mesh Pattern representing who can act is limited by how much of the ledger an agent can read and by human approval, creating auditable authority and reducing mistaken or forged commitments.
Key Data
11,048 tasks recorded across two deployments, with 983 completed — a ~93.9% completion rate from ledger-driven task handoffs.
2About 258,000 ledger events captured during daily use, providing a dense audit trail of agent actions and reads.
3Over 3,000 agent records and 404 teams accumulated between March and July 2026, showing practical scale for personal multi-agent operation.
Implications
Product and platform engineers building agent ecosystems will get a practical pattern for durable, auditable communication that avoids brittle ad-hoc transfers. Technical leaders and SRE/ops teams can use the ledger model to enforce who may commit actions and to inspect what agents actually saw when they made decisions. Researchers studying multi-agent trust and agent-to-agent evaluation get a concrete protocol and field measurements to compare against ephemeral, agent-only designs. For researchers and engineers exploring governance and coordination, see Consensus-Based Decision Pattern and Semantic Capability Matching Pattern.
Need expert guidance?We can help implement this
Ready to evaluate your AI agents?
Learn how ReputAgent helps teams build trustworthy AI through systematic evaluation.
Learn MoreYes, But...
The implementation covers the personal-side ledger and embassy; cross-civilization reachability (public ingress and outage handling) remains an operational challenge. Adversarial inbound messages, forged commitments, and Sybil-style attacks are assumed threats, so binding actions still require human gates and careful deployment. The evaluation is from a single sovereign’s deployments — inter-organization scaling and varied threat environments need more study.
Methodology & More
People acting as the transport layer between AI assistants cause repeated, lossy handoffs: installation guides miss constraints, context drifts, and multiple human round trips are required to reconcile differences. The proposed solution treats a person plus their persistent record as the natural communication unit, called a "civilization." Each civilization runs a lightweight, always-on gateway (the embassy) that accepts inbound items, appends them to an append-only ledger with timestamps and receipts, and lets any agent later claim and act on those items based on the ledger state. Authority to act is tied to what an agent can read from the ledger: agents have graded representation scope (task, project, global), and exercised authority requires human approval and behavioral trust signals in addition to scope. To avoid accidental priority given to earliest-arriving messages, the protocol seals inbound items so order alone does not confer authority; legitimate hierarchy comes from explicit memory scope and model strength instead. A working implementation showed agents can resume one another’s tasks from ledger state, verification agents can catch poisoned inputs by refutation defaults, and structured summaries reduce accidental judgment drift. The trade-offs are operational: reachability, exposure to web threats, and adversarial messaging require careful gateway design and human arbitration for critical commitments. Reflection Pattern
Need expert guidance?We can help implement this
Credibility Assessment:
Single-author arXiv paper with no affiliation or recognizable author reputation—minimal identifiable credibility signals.