Odd Lots Podcast Summary

Episode Overview

Episode Title: Goldman CIO Marco Argenti on the Warp-Speed Improvements in AI
Date: March 30, 2026
Hosts: Joe Weisenthal & Tracy Alloway (Bloomberg)
Guest: Marco Argenti, Chief Information Officer, Goldman Sachs

Theme:
This episode examines how artificial intelligence (AI) is rapidly transforming finance, software development, and enterprise productivity, with a specific focus on Goldman Sachs’ real-world use cases. Marco Argenti details the remarkable changes within Goldman over the past two years—from the experimentation phase to full-scale, agent-powered implementation—and discusses the broader implications for technology, business operations, workforce composition, and regulatory practices.

Key Discussion Points and Insights

1. From Experimentation to Full Adoption

• Not Just a Drill—AI Goes Critical
  • AI’s role in business has evolved from experimental chatbots and productivity tools to mission-critical infrastructure.
  • Quote: “This is not the drill, this is real. It's not the age of experimentation anymore. This is a tool that now can do a lot for you.”
    — Marco Argenti [06:57]
• Six months ago, terms like "agents" were barely in the conversation. Now, AI agents are ubiquitous across enterprise workflows.

2. Real-world AI Deployments at Goldman Sachs

• GSAI Assistant
  • Rolled out to 47,000 employees; used daily, often multiple times per person.
  • Provides rapid, complex research by fetching and synthesizing both internal and external data.
  • Example: Analyzing impacts of geopolitical events or interest rate changes on portfolios in minutes, not days.
• Legend AI Lakehouse
  • Bridges data sources and GSAI, allowing easy, curated data linkage and better AI answers.
  • Data quality and curation are key to high-impact AI.
• Agentic Developer Tools
  • Developers use tools like Cloud Code, Devin, and GitHub Copilot.
  • Paradigm shift: Developers are now planners and supervisors, not just coders.
  • Quote: “The most important thing for a developer today is to be able to explain things rather than jumping in.”
    — Marco Argenti [11:27]
• Productivity Measurement
  • Output, timelines, and quality are key metrics.
  • Projects are now consistently ahead of schedule; optionality boosts business growth.
  • Argenti: AI doesn’t reduce developer headcount, but raises output and innovation.
  • Quote: “You have the optionality of say, now I have 120% of my capacity...Do I want to do 130% more? Great. If I don't, I have the option to reduce.”
    — Marco Argenti [13:29]

3. AI Disruption of Legacy Software Providers

• Cycle of software disruption is now much faster; processes that can change will transform, others (like general ledger/accounting) may be resistant/slow to change.
• Software procurement shifts:
  • “Buy vs. build” pendulum swings toward "build" for small, simple apps due to vastly reduced development cost.
  • Larger, highly complex applications will remain with vendors for now due to scale and integration demands.
  • Quote: “Now I'm starting to see people coming to me and say, by the way, I had some time...here is a perfectly working application. So...the cost...has gone down quite dramatically.”
    — Marco Argenti [21:46]
• Contracts with some third-party software providers have already been terminated in favor of AI-powered in-house tools.

4. Forward Deployed Engineers and Direct Model Providers

• Forward Deployed Engineers: AI model creators embedded within client orgs (like Goldman) to speed innovation and eliminate third-party bottlenecks.
  • Product builders, not just solution architects.
  • Quote: "When things change so much and so rapidly, you kind of want to go to the origin of who produces this new thing."
    — Marco Argenti [23:18]

5. Integration and Productivity Surges

• AI accelerates integration but system-of-record platforms and deep integrations remain critical.
• “Vibe coding”: Employees rapidly develop complex tools or migrate cloud apps during meetings or over a weekend—unimaginable two years ago.
• Examples:
  • Full migration of legacy apps to cloud in hours.
  • Creation of travel assistant automating scheduling and rebooking.

6. AI Compute Budgeting, Token Economics & Resource Optimization

• Organizations now face "token sticker shock"; token allocation and monitoring is crucial.
  • Centralized access and a model gateway optimizes cost/quality.
  • Quote: “There’s going to be a token sticker shock for CFOs... they're going to start seeing bills they absolutely did not expect.”
    — Marco Argenti [27:54]
• Central AI teams manage which queries go to which models to maximize the Pareto frontier of quality and price.
• Quote: "My philosophy is to try to isolate the developer or the user from the token anxiety."
  — Marco Argenti [31:13]
• Expect token unit prices to fall, but total token spend will rise as organizational consumption soars.

7. Security, Regulation, and Data Barriers

• Employees can’t freely install cutting-edge models; strict controls are required given banking regulation.
• Goldman incorporates advanced agentic features internally rather than using public agentic platforms “as is.”
• Speed vs. Velocity:
  • Goldman optimizes for velocity—a sustained, secure pace—rather than just "speed."
  • Quote: "There's a difference between speed and velocity... Speed is a sprint, but at some point, you hit a wall..."
    — Marco Argenti [40:04]
• Regulatory Dialogues
  • Decades of neural network risk management provide a framework for governing current LLMs and agent tools.
  • Key controls: Inventory, risk tiering, human supervision, zero-trust code auditing, continuous integration.
  • Regulators focused on risk controls, inventory, and transparency—not perfect model “explainability.”

8. Information Asymmetry and AI’s Limits

• Classic source of banking profit (access to unique info) may be eroded by democratized AI, but Goldman believes they offer a “final 10%” in edge through data scale, cross-market visibility, and expertise.
  • Quote: "For sophisticated clients, that 10% is really where the money is..."
    — Marco Argenti [46:16]
• Strict info barriers enforced via badge-based access—preventing AI-powered data leakage between regulated business lines. Full-featured, secure platform took two years to build.

9. Changing Nature of Developer & Employee Talent

• AI shifts required skills: ability to explain/ideate, delegate to agents, and supervise output.
  • Management skills (explain, delegate, supervise) are now core for all, not just managers.
  • Quote: "AI is kind of turning everybody a little bit into a manager.”
    — Marco Argenti [51:12]
• Work is more collaborative and dynamic; culture is influenced by "forward deployment" and peer pressure to leverage AI.

10. Work Satisfaction and Burnout in the AI Era

• Some developers report "slot machine" fatigue from constant prompting and iteration, risking burnout.
• Over time, AI becomes seen as just another professional tool; initial excitement gives way to more stable, focused use.
• Mundane, repetitive development tasks are now eliminated, allowing focus on higher-level problem-solving and planning.
• Quote: "I think overall nobody really likes to have that toil and that mechanical work. And I'm actually quite happy that people are gonna spend maybe initially more time because they're excited, but on things that are enjoying rather than things that... they dread.”
  — Marco Argenti [56:01]

Notable Quotes & Memorable Moments

• “This is not the drill, this is real. It's not the age of experimentation anymore. This is a tool that now can do a lot for you.”
  — Marco Argenti [06:57]
• “...I had some time, you know, this weekend and here is a perfectly working application.”
  — Marco Argenti [21:46]
• “There’s going to be a token sticker shock for CFOs... they're going to start seeing bills they absolutely did not expect.”
  — Marco Argenti [27:54]
• “There's a difference between speed and velocity... Speed is a sprint, but at some point, you hit a wall...”
  — Marco Argenti [40:04]
• “AI is kind of turning everybody a little bit into a manager.”
  — Marco Argenti [51:12]
• “For sophisticated clients, that 10% is really where the money is...”
  — Marco Argenti [46:16]

Timestamps for Important Topics

• [05:35] – Marco Argenti returns; summary of AI’s acceleration since 2024
• [06:57] – “This is real, not a drill”: The end of AI experimentation phase
• [08:30] – GSAI Assistant and advanced AI use cases at Goldman
• [12:40] – Productivity and output metrics from developer AI tools
• [17:32] – AI’s impact on legacy software providers, buy-vs-build shift
• [23:04] – What is a “forward deployed engineer” in 2026?
• [25:21] – AI and software integration advances
• [26:46] – Novel use cases built rapidly with AI: “vibe coding” examples
• [27:36] – Computing budgets and token allocation in enterprise AI
• [32:57] – Future of token costs and resource optimization strategies
• [37:16] – Security, agentic AI, and balancing innovation & controls
• [42:01] – Regulatory discussions, model risk management, and compliance
• [46:00] – Information asymmetry, profitability, and what’s left for banks post-AI
• [48:19] – Info barriers to prevent AI-powered data leakage across business silos
• [50:15] – How AI is changing the profile of ideal hires
• [52:14] – Developer work satisfaction, burnout, and the “slot machine” effect
• [56:01] – Removing toil from developer work and increasing job satisfaction

Tone & Language

• The episode is lively, candid, and approachable, with hosts pushing Marco for concrete answers and real-world examples.
• Argenti’s tone balances technical detail, strategic vision, and humor, clearly relishing both the engineering and cultural challenges AI imposes.

Final Thoughts

This conversation provides an unusually transparent, inside look at how a global financial powerhouse is actually building, governing, and leveraging AI—and how quickly those strategies are evolving. If you want to understand enterprise-scale AI’s real impact—not just the hype—this is a must-listen episode.

For deeper dives:

• Listen at [08:30] for practical workflow examples
• [27:36] for internal AI budgeting
• [42:01] for regulatory & compliance frameworks
• [50:15] for the evolving workforce dynamics