Goldman Sachs gave 10,000 employees, roughly a quarter of its entire workforce, secret access to an internal AI tool for over a year before anyone outside the company knew it existed.
That quiet pilot phase is worth noting on its own. While competitors were announcing AI initiatives in press releases, Goldman spent a year running a closed test with a meaningful chunk of its actual staff before deciding it was ready for a real rollout.
When they did launch publicly in mid-2025, they didn't ease in. The GS AI Assistant went to all 46,500-plus employees worldwide, firewalled entirely behind Goldman's own infrastructure specifically so no sensitive client or trading data could ever leak into an external model. CIO Marco Argenti, hired from Amazon, built it as a simple chat interface, but with something more sophisticated running underneath.
Here's the architecture decision that actually matters. Goldman built a multi-model system with role-based behavior, meaning a banker's version of the tool is tuned differently than a research analyst's or a developer's, and access to the most capable, most expensive models is reserved specifically for genuinely hard problems, while routine tasks run on smaller, faster models. That's a deliberate cost and capability tradeoff most companies skip entirely by giving everyone the same generic access.
The results reported internally: adoption crossed 𝟱𝟬% 𝗼𝗳 𝗲𝗹𝗶𝗴𝗶𝗯𝗹𝗲 𝗲𝗺𝗽𝗹𝗼𝘆𝗲𝗲𝘀, with a stated goal of 100% by 2026. Developers using the AI Developer Copilot saw a 𝟮𝟬% 𝗽𝗿𝗼𝗱𝘂𝗰𝘁𝗶𝘃𝗶𝘁𝘆 𝗶𝗻𝗰𝗿𝗲𝗮𝘀𝗲 and a 𝟭𝟱% 𝗿𝗲𝗱𝘂𝗰𝘁𝗶𝗼𝗻 𝗶𝗻 𝗽𝗼𝘀𝘁-𝗿𝗲𝗹𝗲𝗮𝘀𝗲 𝗯𝘂𝗴𝘀. For investment bankers, internal estimates suggest AI-assisted document drafting and financial modeling can cut M&A deal preparation time by up to 𝟰𝟬%. Their Legend Copilot tool lets analysts run natural language searches across Goldman's proprietary data repositories instead of manually digging through internal systems.
Here's the honest caveat worth including. Goldman's own equity strategy team built a stock portfolio called GSXUPROD, specifically betting on companies positioned to benefit most from AI-driven productivity gains. As of their 2026 report, that portfolio was actually underperforming the broader market year to date, even excluding the largest tech stocks, though Goldman's analysts still argue its long-term earnings growth potential exceeds major indices.
Here's the lesson for product leaders.
Even the bank betting the most confidently on AI's future value hasn't seen that thesis play out cleanly in the market yet. Internal productivity gains and external investment returns are two completely different measures of whether an AI bet is working, and it's worth being honest that only one of them is showing up clearly right now.
Which Global 2000 company should I break down next.
ArtificialIntelligence #ProductLeadership #Banking #EnterpriseAI #AIStrategy
[comment, follow, save, share, repost]
Top comments (0)