How much payment authority are people giving their agents in production?

What I've seen from those who have dared to deploy agents with spending/financial capabilities, there seems to be three distinct comfort levels in practice. Most, as expected (still early days), are at the query and recommend stage, agents surface options, humans authorize every transaction. Basically a well dressed dashboard. Those that are actually shipping payments tend to be running hard per transaction caps with daily limits and human review at the end of the day. Lastly, an even smaller group has agents with broader payment authority in a specific domain, buying their own compute credits, paying per call APIs, and very rarely opening trading positions (I see a lot of talk about this, but no so much in production). These are usually builders more familiar with agentic payments, and have been running their agents for months and built up a trust profile slowly with time. Most of the content about agentic payments talks about that third groups as if it's the norm. From what I've seen, most production deployments are in the first and cautiously moving towards the second. Don't think we're at the third just yet.