Goldman Sachs’ tech boss says tracking individual AI usage isn’t useful. He just watches how fast his 12,000 engineers move from idea to production

Goldman Sachs CIO Marco Argenti said the bank is measuring AI productivity by how quickly engineering teams move from idea to prototype, rather than by individual employee AI usage. Goldman says its 12,000 engineers are already using AI through tools like GS AI Platform, an internal ChatGPT-style system, and Legend, which helps employees search internal files with natural language. The article is mainly a qualitative update on Goldman’s AI adoption strategy and is unlikely to move the stock materially.

Analysis

The market is still underestimating the second-order implication of this message: AI monetization in large enterprises is moving from individual productivity anecdotes to workflow-cycle compression. That is a bigger operating leverage story for large banks and consultancies than simple license-count growth, because the real payoff shows up when teams ship faster with the same headcount, not when usage dashboards spike. For GS, that argues for a gradual margin tailwind over the next 4-8 quarters if management can convert faster prototyping into fewer handoffs, lower rework, and better client-facing product velocity.

The more interesting competitive effect is that this widens the gap between firms that can safely operationalize proprietary data and firms stuck in pilot mode. Goldman’s security wrapper and internal search stack create a moat around workflow adoption that generic public-model users do not have, which should support continued outperformance in internal automation versus peers with weaker data governance. For GOOGL, the read-through is less about immediate enterprise revenue and more about durable demand for controlled AI infrastructure; the enterprise buyer increasingly wants governed workflows, not just model access.

The contrarian risk is that leadership overstates productivity gains before they are visible in revenue or headcount reduction. In the next 1-2 quarters, the data may look better qualitatively than quantitatively, and the market could punish any AI narrative that fails to show up in operating expense leverage or faster product releases. ACN is the most vulnerable name here: the more clients believe internal teams can "3D print" software, the more pressure falls on lower-value transformation work unless Accenture can prove it is the one enabling those gains, not being disintermediated by them.

Key Decisions for Investors

• Long GS vs. short ACN as a 3-6 month pair: GS benefits from internal AI compounding and governance credibility, while ACN faces margin pressure if clients insource more build work; target 8-12% relative outperformance with a tight stop if ACN re-accelerates bookings.
• Buy GS calls on any post-earnings dip over the next 1-2 quarters: use 3-6 month tenor to express upside from workflow efficiency translating into operating leverage; risk/reward favors limited premium outlay versus waiting for clean margin proof.
• Add GOOGL on weakness as an enterprise AI infrastructure beneficiary: the optionality is not consumer usage, but governed deployment at scale; 6-12 month horizon, with downside limited by cloud and model monetization even if headline AI adoption slows.
• Avoid chasing ACN on AI-fluency messaging until revenue conversion is clearer: if client budgets shift from consulting-heavy implementations to internal AI-assisted delivery, near-term multiple support may compress before any upside from advisory demand appears.