JPMorganChase brings Claude beyond engineering into product, risk and operations work

Product planning, risk assessments and infrastructure operations are becoming part of JPMorganChase’s wider use of Claude—not just software engineering. In a November 9 update, Anthropic said the bank is bringing Claude to more employees beyond engineers. The rollout pairs broader business use with staged access, internal model approvals and controls over changes that reach production.

It says work in the Commercial and Investment Bank is reaching production faster, with the resulting capacity redirected toward business priorities.

From business requirements to engineering work

Product teams use Claude to turn requirements into engineering backlogs—the lists of work that development teams need to tackle. Other teams use Claude Code for risk and controls assessments and program planning. Those uses take the tool into work surrounding software delivery, including deciding what needs to be built and organizing how a program will proceed.

The account also describes a team structure in Asset & Wealth Management that brings people with business expertise together with technology and product specialists. These small teams work on customized client investment solutions. Anthropic says Claude Code is now part of their engineering workflow, and that the teams are delivering faster while testing changes more rigorously before clients receive them.

Agents for maintenance and operations

Engineering teams are also using Claude to modernize legacy systems and build their own agents—software that carries out tasks. Anthropic gives two examples: agents that review proposed code changes, known as pull requests, and self-healing agents. Engineers are sharing approaches that work, making the deployment a source of internally developed workflows as well as access to the underlying tool.

Wider access, separate approval

The bank runs Claude Code in controlled environments on its own infrastructure, including Devspace, its containerized development platform. Containers provide separated environments for development work. Anthropic describes the intent as preventing agents from gaining direct access to employee credentials or internal systems; that is the stated purpose of the setup, not a claim that every possible access risk has been eliminated.

Access also widens in stages. Before employees can use a new Claude model, it must pass JPMorganChase’s own approval process.

For code headed into production, Anthropic says thousands of automated tests and approval gates check every change. The enforced pass-or-halt decision is part of that delivery process, rather than something Claude can grant itself. People remain accountable for every production change, even as more teams use the tool to prepare, build and investigate work.