Goldman Sachs restricts Claude Code and GitHub Copilot to engineers

📅 2 November 2026 • The Decoder / Bloomberg

Goldman Sachs has drawn a clear line around who can use advanced AI coding tools inside the bank. According to The Decoder, citing Bloomberg, internal rules introduced in October 2026 keep Claude Code and GitHub Copilot available only to engineers. Traders and bankers, who previously had access to both tools, have lost it. The tools in question write, refactor and run code, and their use had spread well beyond engineering teams, into parts of the business that have no software development function at all.

The restriction reverses months of open usage. Before the new rules, non-technical staff had used these tools for months, including for monitoring trading flows and for generating content. In practice the tools had become general-purpose work assistants for employees far from software development — used exactly for the kind of quick, self-serve automation that made agentic coding tools popular outside engineering departments in the first place. That usage had moved past any pilot stage — the tools sat inside daily workflows.

The turnaround is pointed, because the bank promoted the opposite approach this summer. In July 2026, Goldman Sachs CIO Marco Argenti publicly promoted vibe coding for non-technical employees — the workflow in which an employee turns an idea into a fast prototype with an AI tool and then hands the result to an engineer. That public position framed AI coding tools as a way for any employee to automate parts of their own work. By October 2026, access for that same group of employees has been withdrawn.

The official framing for the restriction is productivity and risk management. According to the reporting, the experimentation phase with these tools is considered finished, and access stays with those who are best suited to long-term work with them. The Decoder additionally notes the internal reaction — the decision caused frustration among employees inside the bank, for whom the tools had already become part of the daily working routine.

For enterprise AI governance, the case is a clean data point. Access to agentic tools is being aligned with role and with long-term responsibility for what those tools produce, rather than with curiosity or short-term experimentation. The sequence at Goldman Sachs — broad experimental access in July, public promotion of vibe coding by the CIO, role-based restriction in October — shows how a large organization recalibrates its AI tooling once the pilot phase ends. Other banks and large companies running comparable programs now have a reference case for what that consolidation looks like in practice.

IMPACT. Who is affected — traders and bankers at Goldman Sachs who lose access to Claude Code and GitHub Copilot, engineering teams that keep them, and AI governance functions at other large firms running similar programs. What changes in processes — access to agentic coding tools is set by role and long-term ownership rather than open experimentation, and tooling granted during a pilot phase gets consolidated once that phase ends. What to watch — whether other banks and large enterprises adopt the same role-based access model for agentic AI tools.

Sources: the-decoder.com, bloomberg.com

Continue the conversation: https://t.me/AIAutomationArchitecture · Bluesky