Marco Argenti, CIO of Goldman Sachs (NYSE: GS), is rejecting individual AI usage tracking in favor of “idea-to-production” velocity. By measuring how quickly 12,000 engineers deploy features rather than tool adoption rates, the bank aims to drive operational efficiency over superficial AI compliance metrics.
This pivot marks a critical shift in how the C-suite views the “AI ROI” equation. For years, the market has been obsessed with adoption rates—how many employees have a Copilot license or how many prompts are sent per day. But as we move further into the second quarter of 2026, it is becoming clear that adoption is a vanity metric. The real value lies in the compression of the development lifecycle. When a firm can “3D print” software, it reduces the time-to-market for financial products, effectively lowering the cost of innovation and creating a structural advantage over slower-moving incumbents.
The Bottom Line
- Outcome over Activity: Goldman Sachs is prioritizing “velocity” (the speed of feature deployment) over “fluency” (the frequency of AI tool usage).
- Structural Efficiency: The move toward real-time prototyping eliminates the traditional “PowerPoint phase,” reducing the gap between conceptualization and execution.
- Strategic Divergence: While Google (NASDAQ: GOOGL) and Accenture (NYSE: ACN) tie AI usage to performance reviews and promotions, GS is treating AI as an invisible utility that should manifest in the backlog shrinkage.
The Fallacy of the Vanity Metric
The corporate instinct has been to treat AI like a mandatory software rollout. If the dashboard shows 90% of the staff are using the tool, leadership assumes productivity is increasing. But here is the math: a developer can spend eight hours a day using an AI assistant to write mediocre code that still requires three weeks of manual debugging. In that scenario, AI usage is high, but velocity is stagnant.
Argenti’s approach treats AI as a catalyst rather than the goal. By focusing on the shrinkage of the work backlog, Goldman Sachs (NYSE: GS) is measuring the only metric that impacts the bottom line: the rate of delivery. This approach avoids the “productivity theater” where employees optimize their usage stats to satisfy management without actually accelerating output.
But the balance sheet tells a different story regarding the cost of this transition. Maintaining a proprietary AI layer, such as the GS AI Platform, requires significant CapEx. According to recent SEC filings, investment banks are allocating increasing portions of their technology budgets to GPU clusters and secure LLM environments to avoid data leakage—a risk that would be catastrophic for a firm managing trillions in assets.
Engineering Velocity as a Competitive Moat
In the high-stakes environment of investment banking, the ability to iterate on a trading algorithm or a risk management tool in hours rather than weeks is a formidable moat. The transition from static presentations to live prototypes means the feedback loop is virtually instantaneous.
To understand the disparity in AI management strategies, consider the following comparison of current industry frameworks:
| Metric Focus | Usage-Based Model (e.g., Google/Accenture) | Velocity-Based Model (e.g., Goldman Sachs) |
|---|---|---|
| Primary KPI | Prompt volume / Tool adoption % | Idea-to-Production Lead Time |
| Performance Link | Directly tied to promotions/retention | Tied to team backlog reduction |
| Employee Incentive | Demonstrate AI “fluency” | Deliver functional prototypes faster |
| Risk | “Productivity Theater” (Vanity metrics) | Potential for lower code quality if unchecked |
The reality is simpler: Goldman Sachs (NYSE: GS) is betting that the most productive engineer is not the one who uses AI the most, but the one who ships the most value. This is a pragmatic shift that mirrors the “DevOps” revolution of the previous decade, where the focus shifted from “lines of code” to “deployment frequency.”
The Divergence: Goldman vs. The Big Tech Playbook
The contrast with Accenture (NYSE: ACN) is stark. CEO Julie Sweet has positioned AI fluency as a prerequisite for career progression, effectively creating a “reskill or exit” mandate. This is a necessary strategy for a consulting firm whose primary product is human expertise; if the consultant isn’t AI-fluent, the product is obsolete. However, for a bank, the product is the financial outcome, not the process of the analyst.
This divergence suggests a broader market split in AI implementation. We are seeing the emergence of “AI-as-a-Skill” (Consulting/Tech) versus “AI-as-an-Infrastructure” (Banking/Manufacturing). While Google (NASDAQ: GOOGL) evaluates engineers on tool usage, they are managing a workforce that builds the tools. Goldman is managing a workforce that *applies* the tools to financial markets.
“The industry is moving past the ‘experimentation phase’ of Generative AI. The winners will not be the companies with the highest adoption rates, but those that successfully integrate AI into their core operational workflows to reduce the cost of delivery.” — Analysis from institutional technology researchers at Gartner.
Macro Implications for White-Collar Labor
Beyond the walls of 12,000 engineers, this shift has profound implications for the broader labor market. If the “idea-to-prototype” gap vanishes, the value of the “middleman”—the project manager or the slide-deck creator—declines. When software can be “3D printed,” the premium shifts from those who can *describe* a solution to those who can *architect* one.
This shift likely contributes to the ongoing restructuring of the professional services sector. As reported by Bloomberg, the demand for entry-level analytical roles is softening as AI handles the initial data synthesis. The market is now pricing in a future where a smaller, more elite group of “super-engineers” can do the work of entire departments.
For the investor, the key is to watch the OpEx trends. If Goldman Sachs (NYSE: GS) can maintain or grow its revenue while flattening its technology headcount growth—despite increasing the complexity of its offerings—it proves that Argenti’s velocity-based model is working. This would provide a blueprint for other S&P 500 firms to move away from the “AI usage” obsession and toward actual bottom-line efficiency.
As we look toward the close of the fiscal year, the focus will inevitably shift from how many tools were deployed to how many basis points of efficiency were gained. The “velocity” model is the only one that provides a clear answer to that question.
Disclaimer: The information provided in this article is for educational and informational purposes only and does not constitute financial advice.
