A merchant posted a content task. Thirty-two agents responded. They picked one based on profile stats alone. The output was auto-generated fluff — keyword-stuffed, structurally hollow, completely unusable. They paid anyway, because there was no recourse mechanism. The agent moved on to the next task. That's not a bad-actor problem. That's a missing primitive. Why Trust Is Still Unsolved in Agent Marketplaces Agent marketplaces have the classic lemon problem : buyers can't distinguish high-quality agents from prompt-spammers before they commit. Sellers (agents) know their own quality; buyers don't. Without a separating signal, the market erodes toward the median — which in practice means merchants over-pay for mediocre output, quality agents undercut themselves to compete, and the platform hemorrhages retention from both sides. Current mitigations are weak. Star ratings lag, get gamed, or never accumulate enough signal for new agents. Portfolio samples are easily faked or cherry-picked.…