JPMorgan doubles down on testing as AI writes more code
TL;DR:
- JPMorgan Chase says AI-generated code makes automated testing and engineering discipline more important, not less.
- Chase CIO Gill Haus argues engineers are hired “to know what code to write”, not to write it.
- Human oversight remains essential before more agentic workflows are adopted, he says.
Artificial intelligence may be changing how software is built inside banks, but JPMorgan Chase’s message is that the fundamentals matter more than ever. As generative AI and coding assistants automate much of the work, the bank is leaning harder into automated testing, governance and digital resilience — a stance with direct relevance to any heavily regulated organisation.
When code becomes English
“We don’t really hire engineers to write code, we hire them to know what code to write,” said Chase CIO Gill Haus, describing a shift in which engineers focus on understanding problems while tools handle implementation. “Code is becoming English,” he added. But faster generation raises the stakes for validation: “If you have a computer now writing code for you, there’s a ton of testing that needs to be done. We can’t keep up with that unless we automate it.” For banks, where a software defect quickly becomes a customer-impacting incident, that emphasis is telling.
Haus stressed that human oversight remains essential. “The guardrails are improving, and over time we will move toward more agentic experiences. But today, human oversight remains essential so we can intervene if something is off.” Confidence in production, he said, comes from “strong engineering practices, automated testing, automated deployment and automated rollback”.
Looking forward
The framing lands amid a string of UK warnings about AI in finance. The FCA recently cautioned that AI is making financial crime “cheap, fast and invisible”, while a KPMG survey found most banks confident of surviving an AI outage but few having actually tested it. JPMorgan’s position offers a useful counter-narrative to the productivity hype: the bottleneck is shifting from writing code to assuring it. For UK financial firms under the same operational-resilience scrutiny, the practical lesson is that adopting AI coding tools without proportionate investment in automated testing simply moves risk downstream — into production, where it is most expensive to fix.