AI for Everyone. AI Coding? Not So Fast.

Banks are getting comfortable with AI that helps employees think, write and analyse. But they are more cautious when AI starts doing things on its own.

Goldman Sachs is a good example.

Earlier this year, Goldman allowed its traders and bankers to use AI coding tools like Claude Code and GitHub Copilot. These tools let people build software simply by describing what they want in plain English, without having to write the code themselves.

Traders started using them to build small applications. Some tracked client trading activity, others analysed market positions or created reports and charts. The idea was simple: a trader could build a working version of something useful and then hand it over to an engineer to make it ready for regular use.

But in October, Goldman changed its policy.

These AI coding tools are now restricted to its roughly 12,000 engineers and its specialist staff who work across trading and technology, known as “strats”. Most traders and bankers can no longer use them.

Goldman’s internal AI assistant remains available to everyone. Employees can still use AI to ask questions, analyse information and draft documents. What has changed is who gets to use AI to build and run software. Goldman hasn’t explained the decision in much detail. Its spokesperson said the bank initially makes tools available more widely to understand how people use them, then reviews their benefits and risks before deciding who should continue using them.

The concern is understandable.

In banking, software that handles trades or client information normally goes through several checks. Engineers test it, security teams review it, and someone is responsible if something goes wrong. When anyone can build software using AI, some of these checks can get missed. The person building it may not fully understand the code, how it handles sensitive information or what happens when it produces a wrong result.

Goldman isn’t saying bankers shouldn’t use AI. It is drawing a line between using AI to help with work and allowing AI to create software that does the work.

Other banks are thinking along similar lines. JPMorgan CEO Jamie Dimon said during the October earnings call that the bank would not allow outside AI agents into its systems.

I think this raises an important question for every organisation adopting AI.

Until now, much of the focus has been on getting more employees to use AI.

The next challenge is deciding who should be allowed to let AI take action on their behalf, and under what conditions.

References:

• Finextra — Goldman bans traders and bankers from using AI coding tools
• Quartz — Goldman Sachs restricts AI coding tools for traders and bankers
• The State of AI — Goldman Sachs, Claude Code, traders and bankers