Marco Argenti is chief information officer of Goldman Sachs, which he joined in 2019 after several years at Amazon Web Services. In May 2025 he was a guest of Brian Sozzi, executive editor of Yahoo Finance, on the Opening Bid podcast, and Yahoo Finance published an account of the conversation on 23 May 2025. He talked about where AI fits in the history of workplace technology, how widely it had spread inside Goldman Sachs, the stages he sees in adoption, how AI agents might be managed, and what he expects for jobs.
Another transition, like the spreadsheet
Argenti framed AI as one more in a line of technology shifts. He compared it to the change from working before Excel to working after it, and to the arrival of search engines. Each changed how office work was done, and he put AI in the same family of change.
His suggestion for workers was practical. Look at your own job, he said, and work out which parts of it are where you add something distinctive. AI, as he described it, should take on the repetitive tasks so that people can put their time into the work where they add the most.
“It elevates your work rather than doing repetitive tasks.”
His own career has run through an earlier cycle of this kind. He recalled joining AWS in 2013, which he described as close to the beginning of a long boom in cloud computing.
How far AI had spread at Goldman Sachs
Asked about use inside the firm, he estimated that close to two-thirds of the organisation was exposed to an AI tool in some form. He offered the number as his own estimate of the position at that moment.
Adoption, in his account, arrives in waves. The first wave makes existing workflows more efficient. The second integrates AI more deeply into how the organisation works. The third deploys agents for tasks such as analysis and pulling data out of reports.
Managing agents like a team
Argenti’s view of how agents grow borrowed from ordinary management. Agents scale, he said, by handing work to other agents, in the same way that managers delegate to the people who report to them. In that picture, people spend more of their time supervising agents and less doing every step themselves.
He was matter-of-fact about what happens when agents fall short. They have a shelf life, he said. When one underperforms, the choice is to shut it down or retrain it, sending it back to the drawing board. The remark treats agents as something to be monitored, corrected and eventually retired, with clear options when they fail.
Jobs, displacement and pace
On employment, he accepted that some displacement may happen at first. His argument was that the industries around a new technology tend to create many new jobs. He used Goldman Sachs itself as evidence: about 12,000 of its 45,000 people are developers, roles that would not exist without computers. In his view, the workforce rebalances itself over time.
He expects AI to follow the same pattern, with one difference. This time the change carries more velocity than earlier transitions did. The article closed on his picture of people and machines working side by side, each applying what it does best where it matters most.
What it means for leaders
- Argenti’s job audit works at any level. Ask each team to list the parts of its work where it adds distinct value, then point AI at everything else first.
- His three waves give a usable order of work. ConfX’s reading is that many organisations are still in the first wave, and should get their workflows stable and measured before moving to agents.
- If agents are to be managed like staff, someone has to own their performance: who reviews them, what counts as underperforming, and who has the authority to retrain or retire one.
- The developer example is a reminder that new roles tend to appear in places nobody planned for, so workforce plans need room for jobs that do not yet have titles.
- An exposure figure says little about depth of use. For leaders benchmarking against Goldman Sachs, tracking what people actually do with AI tools will tell them more than counting who has access.
