Goldman Sachs quietly hands the boring stuff to Claude Al | Official Website of Louis Velazquez, entrepreneur, finance guy and tech innovator

Wall Street loves a big technology story, but the quiet changes usually matter more for your job and your portfolio.

Goldman Sachs is working with Anthropic to build AI agents that take over painstaking tasks like trade and transaction accounting plus client vetting and onboarding, according to CNBC.

Engineers from Anthropic have been embedded inside the bank for about six months to co‑develop those autonomous agents on top of Claude, Goldman Sachs Chief Information Officer Marco Argenti told CNBC in an exclusive interview.

He described these systems as “digital co‑workers” for roles that are scaled, complex, and very process‑intensive.

Goldman Sachs has integrated Anthropic’s AI assistant Claude to help workers with tedious, data-heavy tasks.

Photo by Cheng Xin on Getty Images

How Goldman quietly put AI assistant Claude on ledger duty

The deployment plan is more specific than a marketing slogan, which is why I think it matters for anyone working in or investing in financial services.

Goldman has been collaborating with Anthropic to develop AI agents that automate a growing set of internal roles, with an initial focus on “managing trades and transactions, as well as client evaluation and onboarding,” CNBC reported after speaking with Argenti.

For roughly half a year, Anthropic engineers have been embedded in Goldman’s teams to build those autonomous agents around Claude, Argenti told CNBC, in comments later echoed by Reuters.

The work centers on agents that can handle trade and transaction accounting, along with client due diligence and onboarding, said Global Banking & Finance Review, citing Argenti’s characterization of those functions as operationally intensive.

Executives expect these agents to “decrease the time needed to finalize essential operational functions,” particularly around reconciling trades and pushing onboarding files over the finish line, CNBC highlighted.

From my perspective, that is the classic back‑office sweet spot for early AI.

The data are structured, the rules are clear, and the cost of delay shows up quickly in both client frustration and operational risk.

Why Goldman Sachs moved Claude from coding to accounting

The most interesting part of this story for me is how Claude got promoted from coding helper to would‑be accounting and compliance co‑worker.

Argenti said Claude “excels at coding” and then asked whether that performance was unique to software or a sign that the model can “logically navigate complex problems in a step‑by‑step manner,” according to CNBC’s account of his comments.

He told CNBC the firm was “surprised” by how strong Claude turned out to be beyond engineering tasks, especially in accounting and compliance functions that require processing large volumes of documents and data while applying rules and judgment.

More Wall Street

Goldman realized the qualities that make Claude good at coding, like working with complex, structured information, “could be applied to tasks across accounting and compliance,” CNBC reported.

Goldman transitioned to Anthropic’s Claude models after successfully using AI coding assistants for engineers, then started exploring whether those models could support other specialized roles inside the bank.

In my own analysis, that progression is exactly what you’d expect.

Once a model proves it can reason through complex code, the leap to reasoning through a reconciliation rule book or onboarding checklist is not huge, as long as it sees the right data and you keep humans in the loop.

Where I expect AI to land first in your bank

Looking across CNBC, Reuters, and industry write‑ups, the first wave of AI inside Goldman fits a pattern I’ve seen in other highly regulated industries.

Here’s how I’d break down the early targets using Goldman as the live case study.

Trade and transaction reconciliation: Agents are designed to help account for trades and transactions, spotting and resolving breaks faster so backlogs shrink and risk drops, CNBC reported.
Know Your Customer screening and client onboarding: The same agents are being tuned to handle client vetting and onboarding, where they can check documents, screen names, and step through workflows that used to require a lot of manual effort, said Global Banking & Finance Review.
Accounting and compliance review: Claude is already being used inside Goldman’s accounting and compliance departments to speed tasks involving “massive amounts of data” without hiring more staff, Argenti told CNBC.

When I put that together, I see AI landing squarely on the parts of white‑collar work that feel most like grind: matching lines, checking boxes, and checking again.

Jobs, headcount, and what Argenti is actually signaling with AI push

If you sit in one of those functions, the natural question is whether Claude is coming for your job.

Argenti said the bank is still in the “early stages” of developing agents based on Claude and plans to introduce them “soon,” without giving a specific timeline, according to his CNBC interview.

Goldman expects the new agents to greatly reduce the time needed to complete key operational tasks and is looking for efficiency gains rather than immediate job cuts.

Executives believe AI will help “accelerate operations and manage future workforce expansion,” suggesting a focus on slowing hiring growth rather than triggering mass layoffs right away, CNBC said.

The goal is to speed up those data‑heavy jobs without adding more manpower, Observer added, highlighting that the longer‑term impact on third‑party providers and internal roles is still unclear.

My read is that this is a classic big‑bank approach to automation.

You keep most of the current team in place, but the next time a manager asks for 10 new analysts, the answer may be three people and an AI agent instead.

Why investors should care about this “boring” bank-embraces-AI story

For investors, the impact here is more likely to show up in margins than in splashy new products.

Goldman’s AI agent push is part of a multiyear reorganization led by CEO David Solomon to center the bank’s operations around generative AI, with the potential to constrain head-count growth over time, CNBC noted.

Anthropic is courting enterprise clients with offerings like Claude Cowork, according to Reuters. This performs computer tasks for office workers, and noted that Goldman’s agents are expected to streamline an expanding range of internal workflows.

Embedding Anthropic engineers and co‑developing systems gives Goldman more control and tailoring than simply buying off‑the‑shelf AI tools, while also tying key operations to a single vendor, Global Banking & Finance Review pointed out.

In my view, that translates into three big investor takeaways.

You get higher efficiency as back‑office tasks move faster, more leverage on future growth because you need fewer new hires, and a new form of vendor and model risk around Anthropic that shareholders have to watch.

How far Claude could spread inside Goldman

Accounting and compliance are only the first proving grounds.

Claude agents could also be used for employee surveillance and for creating investment banking pitchbooks for clients, Observer reported, citing Argenti’s comments about future use cases.

Those possibilities fit what I’ve seen in other large organizations testing AI agents.

Monitoring communications and activity for policy issues is a logical next step once you trust an AI system with sensitive data, and generating first‑draft pitch materials is exactly the kind of repeatable, time‑consuming work that lends itself to automation.

AI agents are already being used or evaluated by 42 percent of financial‑services respondents, according to a recent Nvidia survey, with top workflows including knowledge management, internal process optimization, and customer support automation, Observer reported.

From where I sit, that suggests Goldman is not an outlier.

It is an early example of how a major bank can redesign itself around AI agents that reason, decide, and act inside workflows that used to belong to rows of junior staff.