NEW YORK — The question Wall Street is quietly working through is how much of its own labor a machine can be trusted to do. JPMorgan Chase keeps revising its answer upward. The largest bank in the United States plans to deploy a new generation of artificial intelligence agents later this year that can work on their own for hours at a stretch, rather than the few minutes today’s versions manage, the next step in a drive to run more of the bank without people.
The bank’s technology chief described the shift in an interview reported on Tuesday. The agents are moving from tools that finish a single task to what the industry has started calling digital workers, software that manages a whole workflow across several disconnected programs. Where current agents stay coherent for two or three minutes, the next ones are meant to run for an hour or two. After that, in the bank’s telling, come agents that hold together for days, and then for weeks.
The groundwork is already large. AI coding assistants are now built into the daily work of more than 40,000 of the bank’s developers, who are said to be shipping software about 25 percent faster than they were in 2024. The biggest gains so far have come in software development and back-office processing, the parts of a bank that look most like assembly lines, and they are starting to reach the revenue-generating desks as well. All of it sits inside the roughly $17.5 billion technology budget and the AI-first blueprint the bank laid out a year ago, when it began describing itself as a megabank built around the technology rather than merely using it.
What the bank is less eager to dwell on is that the agents it is promising are not yet safe to use. By its own account, long-running agents are not ready for corporate deployment because of security concerns, the obvious one being that a piece of software left to act autonomously for hours across a bank’s systems is a hazard as much as a convenience. The pitch, as so often with this technology, is running well ahead of the safeguards. The capability is being announced before it can be trusted.
Underneath the language of productivity is a plainer word the bank does not use, which is replacement. Across corporate America, a growing share of layoffs is now explained by AI efficiency, and the roles most exposed are precisely the white-collar ones a bank is full of: programmers, accountants, auditors, legal and administrative staff, customer service. Goldman Sachs has estimated that 6 to 7 percent of US workers could lose their jobs to the technology. The automation is arriving among American workers already braced for layoffs, in an economy where the same shift is rearranging office hierarchies.

JPMorgan is not doing anything unusual for its size, which is part of the point. Microsoft has rolled autonomous agents into everyday office software, Google has pushed similar systems into its products, and Amazon has tied tens of thousands of job cuts to its own AI build-out. The agentic turn is an industry-wide bet that the next phase of the technology is not chat but action, software that does the task instead of describing how to do it. The biggest institutions are moving first because they have the most repetitive work to hand off and the deepest budgets to try.
It is also a bet that feeds itself. The promise of cheap, tireless digital labor is one of the engines behind the AI trade that has powered US stocks all year, and every announcement of a more capable agent helps justify the valuations. Whether those agents actually perform at scale, or mainly serve to trim headcount and flatter a quarter, is the question the market has been content to defer. The agent that works reliably for weeks is, for now, a slide in a presentation rather than a thing that exists.
What remains unsettled is whether the longer-running agents prove trustworthy enough to put near a bank’s money, and what becomes of the people whose tasks they are built to absorb. JPMorgan frames the project as efficiency, and on its own terms it is. For the workers who do that processing and coding today, it is something else, and the bank has not said how many of them it expects to still be there once the digital workers it is promising finally arrive.

