JPMorgan Chase is preparing to deploy a new generation of AI agents capable of running autonomously for extended periods, Derek Waldron, the bank’s chief analytics officer, told CNBC in an exclusive interview. Where current AI systems typically carry out a discrete task and stop, the incoming agents will sustain coherent, goal-directed work across multiple software systems for an hour or more without needing a human to intervene. “We’ve entered the era of long-running autonomous agents,” Waldron said, noting that a progression toward sessions lasting days and eventually weeks is the longer-term trajectory.
The deployment is expected before the end of 2026. Waldron acknowledged that security and governance requirements have been the primary brake on enterprise adoption of this class of tool, and that clearing those requirements at a bank the size of JPMorgan – the largest in the United States by assets, with an annual technology budget approaching $20 billion – represents a meaningful signal for the broader industry.
Waldron described the underlying shift in terms of what he called intellectual coherence: the ability of an AI system to hold a complex goal in focus, break it into components, delegate sub-tasks, and track progress over time. Recent advances in model reasoning have made this possible, he said, comparing the new architecture to a team manager rather than a single worker. The agents can also write and execute code, navigate web browsers, and interact directly with desktop applications – capabilities that expand the range of workflows they can handle without handoff.
Revenue, Not Just Cuts
JPMorgan has already recorded measurable commercial impact from earlier AI deployments. In its private banking division, AI systems now run overnight scans of market activity, client portfolios, and research to prepare bankers for the following day. The bank attributes a 20% increase in gross sales to these tools, and Waldron said they could ultimately allow individual relationship managers to serve client books up to 50% larger than currently possible.
The framing Waldron used is notable given how corporate AI adoption is typically discussed. Rather than positioning the technology primarily as a way to reduce headcount and trim costs, he emphasized competitive advantage and revenue expansion as the more durable rationale. “For enterprises to win with AI, it’s not about cutting the maximum number of jobs,” he said. That said, CEO Jamie Dimon has stated publicly that some roles at the bank will be displaced by AI, and the firm has said it is developing retraining and redeployment programs for affected employees.
A Shrinking Moat for Software Vendors
One of the more pointed observations Waldron offered concerns what the AI shift means for traditional enterprise software companies. As JPMorgan’s ability to build capable tools in-house grows, its calculus around purchasing third-party software is changing. The bank is now taking a harder look at whether custom internal development can replace vendor products that previously had no practical alternative.
“The moat around certain types of software companies is most certainly diminished versus where it was in the past,” Waldron said – a direct warning to the established vendors whose pricing power has historically depended on switching costs and the complexity of enterprise integration.
From Chatbots to Workflows
Long-running agents have already surfaced in the developer community – Anthropic’s Claude Code and the open-source OpenClaw tool attracted wide attention earlier this year – but their appearance inside a tightly regulated financial institution signals something different. Banks operate under compliance constraints, audit requirements, and liability frameworks that consumer and developer tools do not face. JPMorgan’s planned rollout suggests the governance infrastructure needed to run autonomous AI inside those constraints is close to production-ready.
The bank’s move fits a broader pattern in enterprise AI adoption, where the question has shifted from whether large organizations will deploy AI to how deeply and how quickly. Waldron’s framing – agents graduating from minutes to hours, then days, then weeks of autonomous operation – suggests JPMorgan is already planning for capability levels that have not yet been demonstrated commercially, treating the current deployment not as an endpoint but as the first step in a longer progression.
