While JPMorgan monitors dashboards tracking the AI activities of thousands of individual employees and Meta installs keystroke-monitoring software on US computers, Goldman Sachs is taking the opposite approach.
In his role as CIO, Marco Argenti, who directs a team of 12,000 engineers, is clearly against individual monitoring and prefers to use velocity to measure how fast developers go from idea to production. It cannot be denied that Argentina’s approach is based on a completely different ideology when it comes to managing AI implementation in a corporate environment.
“If you look at the individual, you’re really missing the forest for the trees,” he told Business Insider. As for the sports comparison, “It’s not just about watching a player and how they move,” he said. Instead, Goldman evaluates teams on how fast they clear their queues, develop features and take their projects from idea to production.
Goldman discovered an important insight: increases in initial token consumption do not immediately correlate with coding output. Argento explored this paradox in engineers using AI to plan, write implementation plans, and write business requirements documents before writing code.
Once the initialization phase is complete, both token usage and coding output are simultaneously accelerated. Knowledge of this limitation allows Goldman to confidently invest in AI infrastructure without considering the initial token spend as waste.
Goldman engineers have gone from skepticism to excitement, Argento said. Instead of PowerPoint presentations and six-page documents requiring imaginative interpretation, developers now come up with working prototypes created in real-time using AI.
“You kind of ‘3D print’ software,” Argenti said. He explained how meeting attendees can request changes that engineers implement immediately.
