Goldman CIO Marco Argenti on the Warp-Speed Improvements in AI

Episode Summary

Executive Summary: The episode argues that AI has moved from experimentation to production, especially inside Goldman Sachs, where it is reshaping workflows, software buying decisions, and developer roles. CIO Marco Argenti says AI now drives measurable output gains, faster delivery, and selective replacement of third-party tools, while token costs, data quality, security, and regulatory controls remain central constraints.

Main Topics: AI has entered the production phase (Priority: 5/5): The hosts and Marco Argenti agree the industry has moved beyond novelty and experimentation into real operational use, with AI now embedded in everyday work. Goldman Sachs’ internal AI platform and adoption (Priority: 5/5): Argenti describes Goldman’s GSAI assistant, broad employee usage, and the firm’s centralized model gateway as the backbone of enterprise AI deployment. Developer productivity and changing job design (Priority: 5/5): AI is shifting developers from hands-on coding toward planning, delegation, supervision, and explanation, with output and delivery speed becoming the main metrics. Build vs. buy and software disruption (Priority: 4/5): AI is making it easier to build small internal applications, leading Goldman to terminate some third-party contracts, while large regulated systems remain harder to replace. Token economics and cost optimization (Priority: 4/5): A major theme is how firms should manage token usage, route requests to the right model, and avoid token anxiety while still controlling costs. Security, info barriers, and regulation (Priority: 4/5): As a regulated bank, Goldman must enforce access controls, model risk management, and human review to ensure AI use remains compliant and secure. Talent, culture, and the future of work (Priority: 3/5): AI is changing the skills Goldman wants: employees must explain, delegate, and supervise, making more people function like managers.

Key Arguments: AI is no longer a toy or an experiment; it is now a real tool that can be used for everyday and even mission-critical work with supervision. Goldman’s GSAI assistant is widely adopted across the firm and is used to answer complex questions that previously took hours, days, or weeks. Data quality is a decisive factor in AI performance; curated, understandable data produces disproportionately better answers. Developers are becoming more like product managers and planners, because AI handles more of the mechanical coding work. The main success metric for AI in a corporation is higher output with shorter timelines and better quality, not necessarily headcount reduction. AI is changing the buy-versus-build equation: small applications are increasingly cheaper and faster to build internally. Large, regulated, enterprise-scale software is harder to replace, but AI can still disrupt software tied closely to changing workflows. Centralized model routing and governance are necessary to control token spend, security, and compliance. Token costs per unit should fall over time, but total token spend will likely rise because usage will expand even faster. Banks still have an edge in complex, high-value advisory work because clients pay for the extra 10% of insight, data, and context beyond what generic AI can provide.

Data Points: ChatGPT launch date: November 2022 - Referenced by the hosts as the starting point of the AI acceleration timeline. GSAI assistant users: 47,000 people - Goldman Sachs employees given access to the internal AI assistant. GSAI usage frequency: Most use it every day; most use it multiple times a day - Argenti describes broad and repeated adoption across the firm. Prompt volume: Way above 1 million prompts per month - Goldman’s internal AI usage is growing rapidly. Cloud migration project status: Two months ahead of schedule - Example used to show AI improving delivery timelines. AI platform build time: Almost two years - Time required to build Goldman’s GSA platform with controls and info barriers. Contract changes: Contracts already terminated - Argenti confirms Goldman has replaced some third-party software with internally developed AI solutions. Model routing goal: Pareto frontier of quality and cost - Goldman’s model gateway optimizes between performance and expense. Token economics comparison: Token cost per hour should be less than wage per hour - Argenti’s rule of thumb for positive ROI on AI usage. Firm scale: 100+ countries - Goldman’s global footprint contributes to its data and information advantage.

Pivotal Quotes: "This is not the drill. This is real." — Marco Argenti: Argenti emphasizes that AI has moved beyond experimentation into real enterprise deployment. "The most important thing for the developer today is to be able to explain things. Rather than jumping and coding things." — Marco Argenti: He describes how AI is changing the developer role toward planning and supervision. "Clients are really paying us for that extra 10%." — Marco Argenti: Argenti explains why Goldman’s data, relationships, and expertise still matter even as AI improves.

Implications: AI is becoming a core operating layer for large firms, not just a productivity add-on. Winners will combine strong data, governance, and model routing with human oversight, while software vendors tied to static workflows face growing disruption.