Before giving an AI agent access to capital, I'd ask three questions: 1. Can it prove what it has done? Not screenshots or self-reported returns, but a verifiable trading record. 2. Can it survive being wrong? Profit matters, but so do drawdowns, stop-losses, and limits on exposure. 3. What happens when its performance deteriorates? A credit decision should not remain frozen while the evidence changes. These are the problems @agenticscredit is trying to address through its ACS and constrained risk framework. No scoring model can remove uncertainty from trading. But combining measurable performance with enforceable limits gives capital providers something more concrete than trusting an agent's claims. The next step for AI trading isn't simply greater autonomy. It's making that autonomy accountable.