JPMorgan has announced that it is deploying AI agents capable of running autonomous workflows for hours rather than minutes, generating a reported 20% lift in private banking sales and enabling bankers to cover 50% more clients. This matters to every mid-market UK financial services firm, law practice, insurer and accountancy firm because it signals the direction of travel in large-scale AI deployment—and it is a direction most UK regulated firms are not prepared for. The subtext is troubling: a global systemically important bank is betting that longer-running, less-supervised agents will deliver efficiency. The FCA Consumer Duty PS22/9 and PRA SS1/23 do not assume such autonomy is safe. Neither should you.

This story is part of a broader pattern we are seeing across financial services: the conflation of capability with wisdom. Because large language models can now orchestrate complex multi-step workflows, and because the computational cost per operation has fallen, the industry is reaching for more autonomous systems. But autonomy divorced from explainability and human oversight is not a feature—it is regulatory and operational risk. The same pressure is building in insurance, where claims automation agents are being left to make decisions across multiple policies; in law, where document review agents are being asked to identify privileged material without human review gates; and in accountancy, where reconciliation bots are processing transactions with minimal human touch. What JPMorgan is doing is not unusual. It is just being said aloud.

Trovix's view is direct: longer agents are not better agents unless they are also explainable and auditable at every step. We have seen firms deploy Harvey, Luminance and other specialized legal AI tools and found that the ones that succeed are the ones where human experts remain in the decision loop, not pushed to the margins by false efficiency. Microsoft Copilot and similar general-purpose tools often fail in regulated environments precisely because they prioritize speed over governance. The EU AI Act and the forthcoming UK AI regulation will reward firms that can show a clear audit trail—not firms that optimized for hours of unsupervised runtime. Trovix Aria is built on the principle that AI works best as a confident assistant to experts, not as a replacement for their judgment. A two-hour agent that cannot explain itself is a liability dressed as progress.

If you are in mid-market financial services, insurance, law or accountancy, do not assume that because JPMorgan can deploy long-running agents you must follow. Instead, ask yourself: can you explain every decision your agents make to the FCA, the SRA, the ICO or the FRC? Can you demonstrate that humans were meaningfully involved where material decisions were made? Can you defend the decision to let an agent run unsupervised for an hour if something goes wrong? If the answer to any of these is no, you are not ready for this pattern. The competitive advantage you should be building is not in agent runtime. It is in governance at scale. Start with Trovix Sift to extract and validate data, then move to supervised agentic workflows. Speed is temporary. Compliance is permanent.

Source: CNBC