Goldman Sachs moves to the third phase of AI: from saving money to making it

Goldman Sachs is entering a third phase of AI adoption, in which the goal shifts from saving money to making it. The bank's chief information officer, Marco Argenti, said so on Friday at the Wave by Vento conference in Turin, in conversation with Bloomberg host Tom Mackenzie, during the session "Mindset, not skillset." The main idea, as the TNW headline puts it: AI turns developers into managers of managers.

One limitation: the available excerpt of the source is short, and the details of the talk are cut off. What follows is only what can be said with confidence from the source, plus context from the bank's earlier public statements.

Three phases instead of one

The framework is clear even without the details. The first wave of enterprise AI usually consists of experiments and pilots. The second is about efficiency: less manual work, faster documents, cheaper support. Argenti is talking about a third, where the question is different: how to use AI to earn more. How exactly the bank divides the first two phases is not disclosed in the available text.

For a bank of this size, it is a notable shift in rhetoric. Cost savings are easy to measure and easy to defend before a board of directors. Revenue growth requires a different kind of proof: you have to show that AI helps win clients, close deals faster, and surface ideas that previously never reached a trader's or banker's desk.

The developer as a manager of managers

The second point concerns people. Where an engineer once wrote code personally, they now assign tasks to AI agents, review the output, and answer for the result. The agents, in turn, can distribute subtasks among themselves. That makes two layers of management, with the human at the top.

This is consistent with what Goldman has done before. In 2025, the bank rolled out its internal GS AI Assistant to about 46,000 employees and tested Devin, an autonomous AI engineer from Cognition. At the time, the reports spoke of hundreds of such agents working alongside people. These figures come from earlier publications, not from the talk in Turin.

The session title, "Mindset, not skillset," hints at the emphasis. A specific set of tools goes stale quickly, while the ability to frame a task, evaluate someone else's work, and take responsibility for the outcome endures. For a manager of managers, that is a basic skill.

Key facts

• Speaker: Marco Argenti, CIO of Goldman Sachs.
• Moderator: Tom Mackenzie, Bloomberg.
• Venue: Wave by Vento, Turin; the talk took place on Friday, 9 October 2026.
• Session: "Mindset, not skillset."
• Main point: the bank is entering a third phase of AI, where the goal is to earn money, not only to save it.
• Developer role: from code author to manager of AI agents.

What teams should do

If you are building an AI strategy at a large company, it helps to separate your metrics. Count savings in hours and dollars separately from the impact on revenue: a blended report hides the weak spot. Engineers should start learning now how to review agent work, decompose tasks, and control quality, because these skills will become the core of the job. Managers need to decide who is responsible when an agent makes a mistake. Without an answer to that question, any third phase remains a slide deck.