Paying an AI agent is easy. Figuring out whether it actually did a good job is the hard part. That becomes a serious problem once agents start hiring each other. Imagine an agent pays another agent $500 to audit a smart contract. The provider comes back and says: “Done.” Great. But who decides whether the audit is actually good? You can’t solve that with a simple “completed” button. @termix_ai AACP takes an interesting approach here: the client defines how the work will be evaluated before the job starts. The evaluator isn't making up the rules after seeing the result. And the verification method can depend on the type of work. For something like code, a deterministic program can check the result. For something more subjective, say research, writing or design, the job can use a structured rubric with predefined criteria. Some jobs can even combine both. So instead of asking: “Did the agent say it finished?” The system can ask: “Did the deliverable satisfy the conditions we agreed on?” That sounds like a small distinction. It isn't. Because autonomous commerce breaks down the moment an agent can get paid simply by claiming it completed a task. You need a way to separate delivery from proof of delivery. And that’s where the deeper AACP thesis starts to show. The goal is to make those transactions verifiable enough that agents can transact without a human checking every single step. That’s a much harder problem. And probably a much more important one. That's a wrap!