Goldman Sachs AI Strategy: An Operating Model Rebuilt Around AI

Goldman Sachs frames its AI strategy as a new operating model, called OneGS 3.0. It names six workstreams where AI will change how the firm runs, from client onboarding to regulatory reporting. A year after launch, the firm has not published results for any of them.

The ambition is a new operating model. Goldman’s fourth-quarter 2025 earnings presentation describes OneGS 3.0 as a new operating model propelled by artificial intelligence.

The goals are broad. The same presentation lists six of them: enhancing the client experience, driving productivity and efficiency, improving profitability, strengthening resilience and capacity to scale, bolstering risk management and elevating the employee experience.

The focus is six named workstreams. OneGS 3.0 starts with client onboarding and KYC, enterprise risk management, vendor management, lending, regulatory reporting and sales enablement.

Supporting aims on the same slide include automating error-prone workflows and optimizing headcount.

The industry view is pragmatic. The firm’s 2025 Annual Report says that every new technology creates winners and losers. It also says heavier investment in AI requires diligent risk management.

Outside benchmarks place it outside the top ten. The Evident AI Index ranked Goldman Sachs 12th of 50 banks in October 2026, down from 9th a year earlier.

Goldman Sachs has not said publicly what it will not prioritize.

The platform is built in-house. The GS AI Assistant became available across the firm in June 2025. It had first been piloted for more than a year with more than 10,000 employees, according to an internal memo from CIO Marco Argenti reported by Fortune.

The next step is agents built with Anthropic. In February 2026, Argenti said Anthropic engineers had been embedded with Goldman teams for about six months. They were building agents for work such as accounting, compliance checks and client onboarding.

The agents were in early testing, and the firm had not set a deployment date or shared performance numbers, as reported by AI News.

The workforce stance is cautious. In an October 2025 memo, CEO David Solomon, President John Waldron and CFO Denis Coleman said the firm would constrain headcount growth through the end of the year, with a limited reduction in roles, according to Fortune.

A spokesperson said headcount would still rise over 2025. The firm’s 2025 Form 10-K reports a headcount of 47,400 at year end. Argenti has said the agent work is not about replacing people, at least at this stage.

Measurement is by team, not by person. In a May 2026 Fortune interview, Argenti said tracking individual AI use is not useful. He measures how quickly a team can build a feature instead. He oversees about 12,000 engineers, and has argued that firms need to teach AI their company culture, a point covered in What Real AI Leadership Looks Like.

Ownership sits at the top. The firm’s three most senior executives sponsor OneGS 3.0, and Argenti leads the AI platform and the agent work. No AI-specific investment figure has been disclosed.

AI risk has its own committee. Goldman’s 2026 quarterly filings name a Firmwide Artificial Intelligence Risk Committee, co-chaired by the CIO and the chief risk officer. The annual report ties AI investment to diligent risk management. Published responsible AI principles were not found in the sources reviewed.

AI fluency, coded against the Lead with AI Fluency Matrix (September 2026 edition):

• Fluency floor: None declared. The firm aims to accelerate AI adoption across its workforce, but has not set a usage or skills standard.
• Highest level evidenced: Nothing above adoption. The agents are in testing, and no workflow with a published result was found.
• Portfolio design: Not disclosed. The firm measures team speed rather than individual use.

No outcomes have been published. The earnings presentation sets goals for OneGS 3.0 but gives no quantified targets. The sources reviewed contain no AI savings or productivity figures.

Adoption is the clearest number. By September 2026, about 40,000 of the firm’s roughly 46,000 employees used the GS AI Assistant every week, according to TIME.

The stated measure is team speed. Argenti tracks how quickly a team can develop a feature, which shows up as a shrinking work backlog.

The roadmap is described as multiyear. The October 2025 memo called the change a multiyear effort, without dated milestones.

Change log

• June 2025: The GS AI Assistant becomes available firm-wide after a pilot with more than 10,000 employees.
• October 2025: The OneGS 3.0 memo constrains headcount growth and plans limited role cuts.
• January 2026: Six initial workstreams are published with full-year results.
• February 2026: Agent work with Anthropic is disclosed.
• May 2026: The CIO says the firm measures team speed rather than individual AI use.
• October 2026: Evident ranks the firm 12th, down from 9th.

• Disclosure: 3 of 4. Goldman Sachs answers where it is going, where it will focus and much of what it will take. It has not published how it will know AI is working.
• The focus is specific. Six named workstreams make OneGS 3.0 one of the more concrete AI strategies a bank has published.
• The workforce stance is restrained. The firm slowed hiring and planned limited cuts, while headcount still rose in 2025.
• The say-do gap is in results. A year after launch, no outcome for any workstream or agent has been published.
• Watch the agents. The Anthropic-built agents are the first real test of whether OneGS 3.0 changes how work gets done.

What the numbers show: Revenue grew 23% since 2022 with 2% fewer people, putting revenue per employee above $1.2 million. Goldman has not tied any of that gain to AI.

Revenue, headcount and revenue per employee, FY2022 to FY2025

• FY2022: $47.37bn revenue, 48,500 employees, $977k per employee
• FY2023: $46.25bn revenue, 45,300 employees, $1.02M per employee
• FY2024: $53.51bn revenue, 46,500 employees, $1.15M per employee
• FY2025: $58.28bn revenue, 47,400 employees, $1.23M per employee

Change, FY2022 to FY2025: revenue up 23%, headcount down 2%, revenue per employee up 26%.

Notes: Revenue is net revenues, as reported. Headcount is at year-end.

These figures give context on scale and productivity. They do not show that AI caused any change. Sources: Goldman Sachs 10-K filings and Q4 2025 earnings release.

What is Goldman Sachs’ AI strategy?

Goldman Sachs’ AI strategy is OneGS 3.0, a new operating model that the firm describes as propelled by artificial intelligence. It starts with six workstreams: client onboarding and KYC, enterprise risk management, vendor management, lending, regulatory reporting and sales enablement.

‍

Does Goldman Sachs use AI agents?

Yes, in testing. In February 2026, the firm said it was building agents with Anthropic for work such as accounting, compliance checks and client onboarding. It had not set a deployment date or published results.

‍

Is Goldman Sachs cutting jobs because of AI?

Only on a limited scale so far. In October 2025, the firm said it would constrain headcount growth and make a limited reduction in roles, but headcount still rose over 2025 to 47,400. Its CIO has said the agent work is not about replacing people at this stage.

‍

How does Goldman Sachs measure AI?

The firm has not published AI outcomes. Its CIO measures how quickly teams build features, rather than tracking individual AI use.

‍

How is this profile put together?

Lead with AI profiles use public sources only, such as annual reports, earnings materials, filings and on-record interviews, dated at the time of writing. AI fluency is coded against the Lead with AI Fluency Matrix. Company figures are marked as company-claimed unless they have been independently verified.