AI Agents for Everything, P.2

AI Agents for Everything Part 2 — Money and Medicine Reading time: about 11 minutes. Part 2 of 3. Missed Part 1 ? It covers what an agent is and how it differs from a model. Two industries got there first. Not because they're the most adventurous — quite the opposite. Banking and healthcare are among the most regulated, most risk-averse, most paperwork-buried sectors on earth. That last part is exactly why. Both industries employ enormous numbers of highly trained people to do things that don't require their training: filling in forms, checking one system against another, writing up notes, chasing approvals. Agents are unusually good at precisely this. So finance and medicine became the first places where the technology met a problem big enough to justify the risk. Here's what actually happened. Part A — Finance Where the money went first: catching fraud If you want the single clearest example of agents working at scale, it's fraud detection. Roughly 90% of financial institutions now use AI for it — making it the most widely deployed AI application in the entire industry. The reason it worked here first is that fraud detection has a shape agents love. The task is repetitive, the volume is inhuman, the feedback is fast and unambiguous (the transaction was fraudulent or it wasn't), and — crucially — a mistake is recoverable. Block a legitimate card and you annoy someone. Annoying is survivable. That's a Level 3 task on the trust ladder from Part 1, and banks recognized it. The results are striking: JPMorgan Chase reports its systems have prevented an estimated $1.5 billion in fraud losses , with a claimed 98% accuracy rate and — the underrated number — a 20% reduction in false positives . That second figure deserves a moment. False positives are the tax the rest of us pay for fraud prevention: the card declined at the restaurant, the transfer frozen on a Friday afternoon, the twenty minutes on hold proving you are yourself. Cutting those by a fifth is a real…