Goldman Sachs has limited access to advanced AI coding tools, leaving engineers as the only employees who can still use them. Most traders and bankers who had used the software for months can no longer rely on it for their work.
Access shifts to engineering
The tools include Anthropic's Claude Code and Microsoft's GitHub. Bloomberg reported the change, citing internal guidelines from October. The new limit reverses a period in which non-engineering staff used AI coding tools for practical tasks.
Traders and bankers had used the tools to monitor trading flows and create content. Their use suggests that AI coding software had become part of work beyond traditional software development, helping employees try out ways to handle tasks directly.
The restriction means that engineers retain access while most other employees lose it. The source does not specify how many people are affected or whether the firm plans to restore broader access.
A different approach to experimentation
In July, Goldman CIO Marco Argenti praised “vibe coding” by non-technical staff in an interview. He described a process in which a trader or someone in asset management could turn an idea into a quick prototype with AI, then pass it to an engineer to scale for production.
That model gives business staff a way to explore an idea before engineering teams take on the work of developing it further. Goldman’s current restriction appears to move away from that open experimentation approach, at least for the advanced coding tools covered by the change.
The distinction matters: using AI to make a quick prototype is not the same as building a tool ready for production. In the approach Argenti described, engineers still had a role in taking prototypes forward. The new policy concentrates access to the tools with those engineers from the outset.
Productivity and risk management cited
Bloomberg’s report attributes the policy shift to “a variety of factors including productivity and risk management,” according to Goldman spokesperson Nick Carcaterra. He said the experimentation phase is over and the firm is limiting the tools to people “who are best suited to leverage them for the long term.”
Those comments offer the firm’s stated rationale, but they do not detail specific productivity concerns or risks. The report also does not say whether a particular incident prompted the change. The available explanation is broad: Goldman is reassessing who should use the tools as it moves beyond experimentation.
For employees who had incorporated the software into monitoring trading flows or creating content, that change is reportedly frustrating. The shift therefore has both an operational effect—fewer people can use the coding tools—and a cultural one, as staff lose a way to test ideas themselves.
What the change signals
Goldman’s move highlights a tension in workplace AI adoption. Wider access can let employees with business knowledge build quick prototypes, while tighter access may make it easier for a firm to focus tool use among a smaller group it considers suited to long-term use.
The source does not describe what alternatives traders and bankers will have, or how engineers will handle requests for prototypes and other work. For now, the reported policy marks a clear change in direction: advanced AI coding tools that were used by staff across roles are being limited to engineers.