Goldman Sachs restricts AI coding tools to software engineers

A shift in AI strategy

Goldman Sachs has implemented new restrictions on the use of advanced AI coding tools across its organization. The bank is now limiting access to applications like Microsoft's GitHub and Anthropic's Claude Code primarily to its software engineering teams. This decision marks a significant pivot from the firm's previous strategy of encouraging broad AI adoption among all staff members.

As reported by NDTV Profit, the bank spent several months experimenting with these tools across its banking and trading divisions. During this period, employees used the software to perform a variety of tasks, including tracking client trading patterns and analyzing market movements. They also utilized the technology to prepare reports and generate visual content for the firm.

The move comes as the financial sector grapples with how to best integrate artificial intelligence into daily workflows. While the bank initially wanted to see how many different departments could benefit, the results of the testing period led to a more conservative approach. The firm is now focusing on placing these specific tools in the hands of those it deems best suited for long-term use.

The rise and fall of vibe coding

Earlier this year, Goldman Sachs leadership championed a concept known as 'vibe coding.' This practice allows individuals without deep technical expertise, such as traders and bankers, to build basic AI-powered tools. The goal was to let these employees create initial versions of software that could then be handed off to the bank's 12,000 engineers for professional development.

The bank's Chief Information Officer, Marco Argenti, had previously praised this approach as a way to democratize tool creation. By allowing non-technical staff to experiment with the 'vibe' of a tool, the firm hoped to accelerate the development of internal systems. This strategy was intended to bridge the gap between the needs of the front-office staff and the capabilities of the technical teams.

However, the recent restrictions suggest that the bank found the decentralized use of these tools to be problematic. While the idea of empowering every employee to build software is appealing, it creates complexities in oversight. The bank is now moving away from this open model to ensure that high-powered coding applications remain under the control of specialized departments.

Risk management and productivity

The decision to tighten access was driven by a review of how the tools were being used and the associated risks. Goldman Sachs officials noted that they had to assess a variety of factors, specifically highlighting productivity and risk management. The bank is looking for the best use cases across the firm rather than a blanket rollout of every available technology.

One of the primary concerns for a major financial institution is the potential for security breaches or operational errors. When non-technical employees use tools designed for software engineering, there is a higher risk of creating vulnerabilities or producing inaccurate data. A spokesperson for the bank stated they want to put tools in the hands of those best suited to leverage them.

This move reflects a broader trend on Wall Street where firms are trying to balance the high costs of AI with measurable results. While the tools can speed up work, the bank must ensure that the speed does not come at the expense of safety. By restricting access, the firm aims to create a controlled environment where the benefits of AI can be harvested without exposing the company to unnecessary hazards.

Impact on traders and strats

The new restrictions have caused some frustration among employees who had integrated these tools into their daily routines. Traders, in particular, had been using the AI to streamline their workflows and manage complex data sets. Now that their access has been curtailed, they face a return to more traditional methods for tasks like market analysis and report preparation.

Despite the general restrictions, the bank has carved out specific exceptions for certain roles. Specialist employees known as 'strats' will continue to have access to some of these tools. These individuals occupy a unique space in the firm, with work that spans across trading, quantitative research, and software engineering.

The distinction between 'strats' and other employees highlights the difficulty of creating a universal AI policy. Because 'strats' perform tasks that are technically complex, the bank views them as being better equipped to handle the risks of coding tools. This creates a tiered system of AI access based on the technical nature of an employee's specific job responsibilities.

The broader Wall Street context

Goldman Sachs is not alone in navigating the complexities of AI adoption. Other major financial institutions are also struggling to find the right balance between innovation and regulation. For example, JPMorgan Chase & Co. recently advised its employees to match the complexity of the AI model to the specific task, urging them not to use expensive systems for routine work.

The bank continues to invest heavily in the space, even as it tightens internal controls. Goldman Sachs has developed its own in-house AI assistant that utilizes models from major technology companies like OpenAI, Anthropic, and Alphabet. This allows the bank to offer a curated experience to its staff while maintaining a layer of oversight between the user and the underlying technology.

Looking forward, the bank remains optimistic about the growth prospects of the AI industry. It continues to position itself as a leader in the tech space, recently acting as a lead underwriter for major technology listings. While the internal rules for employees are becoming stricter, the firm's external commitment to the AI economy remains a core part of its business strategy.