Goldman Sachs' CIO just told us what 18 months of real AI deployment looks like inside a bank, and the pace of change is faster than the headlines suggest.
The Summary
- Marco Argenti returned to Bloomberg's Odd Lots 18 months after his last interview, and the bank's AI stack has fundamentally changed in that window.
- Agentic platforms like Claude Code are now in production, actively changing how Goldman's developers and engineers work.
- The real story isn't the tools themselves, it's the data infrastructure and regulatory work required to make them useful at scale.
The Signal
When Goldman's CIO last spoke about AI deployment, the bank was still building internal tools and feeling out the territory. Eighteen months later, the conversation has shifted from "can we use this?" to "how do we scale this without breaking everything?" That's the clearest signal yet that enterprise AI has moved from pilot programs to production environments.
The specific mention of agentic platforms like Claude Code is notable. These aren't chat interfaces or code completion tools. They're systems that can take a task, write code, test it, fix it, and ship it with minimal human intervention. For a bank with thousands of developers, this isn't productivity theater. It's a fundamental shift in how software gets built and maintained.
But here's what matters more than the shiny tools: Argenti spent real time on data challenges and regulatory concerns. That's the unsexy infrastructure work that separates companies actually using AI from companies talking about using AI. Financial services data is messy, siloed, and heavily regulated. Getting it clean enough and accessible enough for AI agents to work with it requires the kind of unglamorous plumbing work that doesn't make for good demo videos but determines whether any of this actually works.
The 18-month gap between interviews is itself instructive. In that window, we went from GPT-4 to Claude 3.5 Sonnet to full agentic workflows. Goldman didn't just watch that happen. They rebuilt their deployment strategy mid-flight to keep up.
The Implication
If you're building in the agent economy, watch what Goldman does next with data infrastructure, not what they say about models. The banks that figure out how to make their data AI-ready without compromising security or compliance will have agents that actually work. The ones that bolt AI onto legacy systems will have expensive chatbots. For developers inside these organizations, the question isn't whether AI will change your job, it's whether you're learning to work with agentic tools now or waiting until someone else already has.
Sources: Bloomberg Tech | Bloomberg Tech