Goldman Sachs says AI adoption depends on redesigning workflows for verification

Goldman Sachs is pushing a blunt idea into its AI strategy: the real hurdle is not whether a model can answer questions, but whether a firm can redesign work so machines can execute it with clear inputs, review steps and outputs. In an October 7, 2025 article tied to the Communacopia + Technology conference in San Francisco, Chief Information Officer Marco Argenti and investment banking co-head Jung Min said AI had already changed a few processes with easily assessed outcomes, including computer coding and customer service, but that broader enterprise adoption still depended on making workflows verifiable.

That distinction matters inside Goldman because it shifts AI from a buzzword to a management problem. Argenti said Goldman was looking at every single process in the firm and asking how AI could interact with it. Min argued that many business processes can be broken into a first person doing the work, a second reviewing it, and a third signing off. Goldman framed that structure as a test of whether a task can be “verifiable by an AI,” a standard that would favor work with clear documentation, repeatable checks and provable outputs.

For employees, the message is that AI adoption will advance fastest where the work can be decomposed and checked, not where judgment is vague or approvals are loosely defined. Coding is easier to test than client-facing or decision-heavy work, which helps explain why Goldman sees a wide gap between functions. The near-term winners are likely to be teams that can standardize processes without flattening quality, while the slower adopters will be the groups where one-off judgment still dominates the workflow.

Min said there were “hundreds of use cases” that are just as verifiable as software development, underscoring how broad Goldman thinks the opportunity is. Argenti has been making the same case all year. In March, he discussed how corporations will use AI and what that means for corporate strategy. In June, Goldman said enterprise AI adoption still had to prove itself and overcome corporate inertia, even as usage of major models rose sharply, hyperscaler AI investment was expected to run about double prior forecasts, OpenAI’s weekly average users climbed from about 180 million in March 2024 to more than 800 million by mid-2025, and leading API token prices fell roughly 100x. Goldman also said token consumption on OpenRouter had risen about 75x since March 2024.

By July, Argenti was arguing that agentic AI would reshape workforce dynamics and that the next generation of “AI natives” would matter because experience and sound judgment are not built into generative AI itself. Goldman’s own AI materials say the technology is already surpassing key human benchmarks in reading comprehension, image and speech recognition and language understanding. The bank’s latest point is narrower and more practical: AI will matter most where firms do the unglamorous work of rewriting procedures so a machine can follow them, and a reviewer can still trust the result.