JPMorgan Chase executives discussed the bank’s AI spending during its third-quarter earnings call, as the company missed analyst estimates. Adjusted earnings were $5.57 a share, below the $5.94 estimate, while revenue of $50.9 billion fell short of the $51.2 billion estimate. Commercial and investment banking results also fell sequentially after a strong second quarter.

CFO Jeremy Barnum said the bank aims to limit dependence on any one supplier by staying agnostic about models and the systems used to run them. JPMorgan uses proprietary models as well as open-weight and open-source models, including some run in its data center, to seek suitable capabilities and prices while controlling costs. CEO Jamie Dimon said model-routing tools could direct queries to the best, quickest or cheapest option, since not every query needs the most expensive model.

Barnum said AI is spread across the organization, making its spending and returns difficult to quantify. Earlier estimates included machine learning used in consumer fraud prevention, while questions now focus on generative and agentic AI’s effects on productivity and expenses, especially in software engineering, as well as possible revenue opportunities. He said those benefits were not observable in the third quarter and may take a long time to emerge at a company the size of JPMorgan Chase.

Barnum described experiments to rewrite minor legacy tools in a couple of weeks; they did not work perfectly, but he said they showed potential for faster work and productivity gains. Dimon said JPMorgan uses agents internally and plans to use them for customers, while also building tools to protect customers from rogue agents. He said customers should understand the risks and any transfer of liability if agents are misused.

JPMorgan Chase executives discussed the bank’s AI spending during its third-quarter earnings call, as the company missed analyst estimates. Adjusted earnings were $5.57 a share, below the $5.94 estimate, while revenue of $50.9 billion fell short of the $51.2 billion estimate. Commercial and investment banking results also fell sequentially after a strong second quarter.

CFO Jeremy Barnum said the bank aims to limit dependence on any one supplier by staying agnostic about models and the systems used to run them. JPMorgan uses proprietary models as well as open-weight and open-source models, including some run in its data center, to seek suitable capabilities and prices while controlling costs. CEO Jamie Dimon said model-routing tools could direct queries to the best, quickest or cheapest option, since not every query needs the most expensive model.

Barnum said AI is spread across the organization, making its spending and returns difficult to quantify. Earlier estimates included machine learning used in consumer fraud prevention, while questions now focus on generative and agentic AI’s effects on productivity and expenses, especially in software engineering, as well as possible revenue opportunities. He said those benefits were not observable in the third quarter and may take a long time to emerge at a company the size of JPMorgan Chase.

Barnum described experiments to rewrite minor legacy tools in a couple of weeks; they did not work perfectly, but he said they showed potential for faster work and productivity gains. Dimon said JPMorgan uses agents internally and plans to use them for customers, while also building tools to protect customers from rogue agents. He said customers should understand the risks and any transfer of liability if agents are misused.