Goldman Sachs is taking AI coding tools back from its bankers
Every company that opens AI tools to everyone eventually has to decide what happens after the experiment. Goldman Sachs has made its call, and it will not be the last bank to make it.
For months, Goldman bankers were allowed to use tools such as Anthropic's Claude Code and Microsoft's GitHub on the job. Internal guidance published last month now limits those tools to the firm's engineers, according to a report this week. A source told the reporters that some traders were frustrated, having used AI to monitor client trading flows and to build written and graphic materials.
The reversal is sharper because of what the firm said earlier in the year. In July, chief information officer Marco Argenti talked up vibe coding by non-technical staff, who built solutions with AI tools and then handed them to one of the firm's 12,000 engineers to strengthen. A spokesperson, Nick Carcaterra, explained the new position: "We make certain AI tools available to our people to determine the best use cases across different parts of the firm. After a period of experimentation where we assess a variety of factors including productivity and risk management, we put them in the hands of those who are best suited to leverage them for the long term."
That is a reasonable thing for a bank to do. Code written by people who cannot review it is a real risk in a regulated business, and the cost of tool licences across a firm that size is not small. The same report notes that banks are spending millions on AI tools while weighing cost, productivity and cybersecurity risk, and other large US banks have drawn their own lines on outside AI agents.
The harder question is what happens to the people who just lost the tool. A trader who learned to build a monitoring script has picked up something useful: a habit of breaking a problem into steps, specifying what good looks like and checking the output. Taking the coding tool away does not remove the need for that habit. If the firm wants those use cases to keep coming, it needs a clear route from a banker's idea to an engineer's build, and it needs bankers who can describe a problem well enough for that hand-off to work.
This is where a lot of AI rollouts quietly stall. Access gets granted, then narrowed, and nobody redesigns the workflow in between. Rules about who can use which tool and for what are a governance job, and our AI Governance workshops cover how to set them without killing the experimentation that produced the ideas in the first place.
Sources:
News and insights for innovation, digital transformation, future of work and L&D leaders.
Stay ahead of learning and development, corporate innovation and digital transformation news. Plus the future of work. For leaders in AU, NZ, HK, SG, the US, the UK and Canada.
