JPMorganChase Widens Claude Code Beyond Engineering, and No Agent Signs Off Its Own Work

JPMorganChase is extending Claude Code well beyond its engineering teams, and the way it staged the rollout is the useful part. Anthropic’s customer story describes use in the Commercial and Investment Bank, where work reportedly moves into production faster. Infrastructure and operations teams use it for monitoring, alerting and root cause analysis, product teams turn requirements directly into engineering backlogs, and other groups use it for risk and controls assessments and program planning. In Asset and Wealth Management, small teams build customized client investment solutions.

Governance sits inside the pipeline. Claude Code runs in controlled environments, including Devspace, the firm’s containerized development platform on its own infrastructure. Access widens in stages, each new Claude model must pass an internal approval process, and thousands of automated tests and approval gates check every change, so Claude cannot sign off on its own work. People remain accountable for production changes. The story gives no headcount, cost or productivity figures, so treat the speed claims as vendor-reported.

Balyasny Asset Management, a $38 billion asset manager, drew the same line in September. Its Chief AI Officer, Charlie Flanagan, called safety “a product and operating-model question” rather than a model-selection question, citing approved data boundaries, least-privilege access, tool-level permissions, logging and human review so that capability gains do not automatically expand authority.

The pattern across both: regulated firms ship agentic work by deciding who is accountable and what a change must pass before access widens. For a company planning its own rollout, build the approval gates and the audit trail first. The outcomes follow from there.