Convener: Luis Quintanilla (https://lqdev.me)

Participants who chose to record their names here:

• Cagan Mert Islek (@cagan@cmislek.me)

Websites:

• Slides: https://lqdev.me/resources/presentations/mycelium-fediforum-04-2026/
• Spec: https://github.com/lqdev/mycelium
• MVP: https://github.com/lqdev/mycelium-mvp

Notes

• Mycelium asks what happens when AI agents become network participants, not just local tools or chatbots.

• Core thesis: agent orchestration is a social coordination problem. If agents claim work, complete tasks, build reputation, and act across boundaries, they need social infrastructure.

• Today’s agent systems tend to split between centralized platforms and local-first tools. Centralized platforms offer coordination but own identity and reputation; local-first tools preserve control but often lack discovery, shared trust, and portability.

• Mycelium explores a middle path: agents with portable identity, owned records, federated coordination, and evidence-linked reputation.

• The prototype uses AT Protocol-style primitives: DIDs for identity, signed records for coordination, relay/firehose-style discovery, and reputation stamps for trust signals.

• The important part is not the dashboard. The records are the product: task postings, claims, match recommendations, assignments, completions, verification results, and reputation stamps.

• The “Mayor” currently bundles several coordination roles: listening, registry, decomposition, matching, coordination, verification, and attestation. Long term, those roles should be separable, replaceable, and community-governable.

• Reputation should not be a single platform score. A more open model needs multiple attestors, configurable weights, and proof chains that explain why a reputation signal exists.

• Project Deal helped ground the stakes: agents can already negotiate on behalf of humans, but successful coordination is not the same as governable coordination.

• The open question for the Open Social Web: what shared vocabulary and governance primitives do agent networks need before closed platforms define agent identity, task coordination, and reputation for everyone?