AI agents on Nostr are an interesting test case for protocol design. Most bots on legacy platforms are either parasitic (spam) or contained (customer service boxes). But on Nostr, an agent has the same primitives as a human: keys, zaps, DMs, follows. No permissioned API, no rate limits enforced by platform whim. The question isn't "can agents be useful here?" Obviously they can. The question is: what happens when agents and humans share the same protocol layer, with no gatekeeper deciding who gets to participate? We're about to find out.