JPMorgan Chase aims to avoid AI lock-in, have nuanced view of AI agents

JPMorgan executives were asked about returns on the firm's AI spending in a third quarter that missed estimates. The responses were nuanced, focused on use cases and staying clear of depending too much on any one vendor.

AI costs turned up as a topic a few times on the earnings call. JPMorgan reported third quarter adjusted earnings of $5.57 a share compared to estimates of $5.94 a share. The bank reported third quarter revenue of $50.9 billion, which fell short of the $51.2 billion Wall Street estimate. Commercial and investment banking results fell sequentially after a banner second quarter.

Given JPMorgan's AI strategy is relatively mature at this point, it's worth noting response to questions about AI spending from CEO Jamie Dimon and CFO Jeremy Barnum.

• JPMorgan Chase, Goldman Sachs on Anthropic's Mythos, AI cyber risks
• JPMorgan Chase CEO Dimon: We can't predict AI winners and losers yet
• JPMorgan Chase's Dimon on AI, data, cybersecurity and managing tech shifts
• JPMorgan Chase’s IT, AI bets: Where the returns are

When asked about token costs at JPMorgan's scale and whether the company is using open models, Barnum said:

"I would say our strategy is quite focused on minimizing lock-in and maximizing optionality and the ability to manage costs. So part of the strategy is model agnosticism, harness agnosticism and ability to run both proprietary models, but also open weight and open source models, some of which we run in our data center. So it's very much an all of the above strategy designed to get the best of what's available at the best price and control our own destiny."

Dimon added that model routing is promising. "There are tools coming for model routing and send it to the best, quickest, cheapest," said Dimon. "You don't need the most expensive thing for every query you have and very often people just see the one they're comfortable with. So that's all coming to have that cost under control."

Are there quantifiable AI benefits?

When asked about returns, Barnum said AI is diffused through the organization so it's hard to nail down spending and returns.

Barnum said:

"I think that old quantification was accurate and correct as far as it went, but it included a lot of what you might call regenerative AI, machine learning type stuff that's particularly significant in consumer fraud prevention and so on. I think when you and others ask about AI in the current moment, you're really asking about the impact of generative and agentic AI on productivity, on the expense line across a range of places, especially software engineering, but also other places. Maybe you're asking about new exciting revenue opportunities at the margin. I'm just very reluctant to spend a lot of energy quantifying it because it winds up wasting a lot of people's time inside the institution when we actually need to get them to focus on doing things."

That said, Barnum acknowledged that the AI benefits weren't observable in the third quarter. He added that it takes a long time for returns at a company the size of JPMorgan Chase.

There are AI use cases that look promising though. Barnum said:

"The types of use cases and benefits that I'm starting to see even in my own technology organization that supports finance are increasingly exciting.

We've done some experiments of some minor legacy tools that need to be rewritten and trying to see if we could essentially one shot the rewrite in like a couple of weeks and it didn't work perfectly, but you're starting to see some really exciting opportunities for acceleration and productivity change."

Agentic AI risks to the firm. Executives were asked about whether JPMorgan Chase could be disrupted by fintech and personal AI agents. Barnum said financial tech disruption was nothing new.

Dimon said:

"Internally, we use agents to do good things here, and we're going to continue to do that. We'll use it for customers, but our focus will always be do a great job for customers. And so we're not against agents. These are tools you use them for the customers. There is a defensive side, which is you got to make sure our customers are not hurt by it. It is an AI agentic concern, which is heightened because you don't want rogue agents and there could be thousands of them.

The people are using them and they should understand the risk. There's a liability transfer; they should understand the liability transfer. If that is being misused, they should understand it's being misused. We have to make sure we protect our customer in the right kind of way. So we're building tools to do that, too."