JPMorgan Is Teaching Its AI to Stop Asking Permission

JPMorgan already has one of the largest AI deployments in banking. LLM Suite, its proprietary generative AI platform, gives an estimated 230,000-plus employees access to models from OpenAI and Anthropic for tasks like drafting emails, summarizing research, and building investment-banking presentations that used to take analysts hours - now generated in minutes. This is labour augmentation (which coincidently is also the term that refers to stimulating contractions during childbirth): AI that increases the productivity of existing workers, and changes their responsibilities. 

What this may be a signal of a pivot from labour-augmenting AI to labour-replacing AI. Derek Waldron, JPMorgan's chief analytics officer, described the shift bluntly: today's agents "don't just run for two or three minutes to carry out a goal or some instructions of a human, they can run for an hour or two." He expects that window to keep stretching - to "multiple hours, then days, then weeks" - with agents eventually acting less like a tool and more like a "team manager" capable of delegating pieces of a workflow to other systems on its own.

In private banking, the first real-world version of this looks like an agent that works overnight: reviewing market activity, checking client positions, and digesting research before the trading day starts, so that bankers arrive to a summary instead of a stack of homework. JPMorgan says tools like this have already driven a 20% increase in private banking gross sales, and the bank thinks the model could eventually let individual bankers manage client rosters roughly 50% larger than what's manageable today.

JPMorgan spends close to $20 billion a year on technology, and its executives have been unusually candid about what that buys beyond faster paperwork. Asked directly about the effect on staffing, Dimon didn't dodge it: "I think we'll be hiring more AI people and fewer bankers in certain categories, and it will make them more productive." He's called the current moment "the tip of the iceberg," with the expectation that "every app, every process, every job will be affected."

This doesn’t necessarily mean mass layoffs or a halt to hiring: JPMorgan's own reporting shows operations and support headcount ticking down a few percentage points while client-facing roles have grown, with the bank favoring reskilling and early retirement. Whether that balance holds once agents can run for days at a stretch, rather than minutes, is the open question. An agent that can only draft an email for you is a productivity tool. An agent that can independently analyze a book of business overnight and decide what needs a banker's attention by morning is doing a chunk of what a junior banker used to do at their desk, theoretically allowing junior bankers to hold greater responsibilities within projects. 

The test isn't whether JPMorgan can build agents that run for hours unsupervised - it's whether the bank's risk, compliance, and security functions are comfortable letting them touch client money and client data while doing it. Waldron himself has framed the slower rollout of long-running agents as a governance problem as much as a technical one. If JPMorgan clears that bar this year, expect Goldman, Morgan Stanley, and the rest of the bulge bracket to follow within months - the same pattern that played out with the first round of firmwide AI assistants.