Goldman Sachs Didn't Partner With Anthropic to Write Better Emails

A Second Venture

Two days later, the internal story came into focus. Speaking at Anthropic's financial services event in New York on May 6, 2026, as reported by Fortune, Goldman CIO Marco Argenti described a deployment already underway inside the firm. This one was built around autonomous Claude agents initially focused on trade, transaction management and client workflows.

Argenti framed the work in three sequential waves:

1) Empower the technology team to operate at a fundamentally different pace.

2) Re-imagine operational processes end-to-end.

3) Use AI to make better risk and investment decisions.

"This is the first time that instead of buying infrastructure, you can actually buy intelligence," Argenti said. That goes beyond a productivity upgrade. It is a fundamental architectural shift.

Taken together, the two announcements describe the same thesis arriving from two directions, namely that frontier AI models are becoming operational layers inside modern institutions, not optional productivity add-ons.

A Focus on Information

Wall Street has always fundamentally been an information business. For decades, financial institutions built massive organizational pyramids around the movement, interpretation, packaging, and distribution of information.

Junior analysts summarized data for senior analysts. Teams built investment memos. Research departments generated reports, and advisors translated products into narratives clients could understand. Entire firms were constructed around the idea that information was scarce, expensive, and difficult to synthesize.

AI changes the economics of that equation. Not because it is magical, and not because it is infallible, but because it dramatically reduces the cost of cognition. But the advisory industry probably prefers to pretend that’s not true.

If part of your value proposition depends upon clients believing that financial analysis, portfolio comparison, planning synthesis, or institutional research access are prohibitively difficult, expensive, or opaque, you should probably pay attention to technologies that compress those barriers toward zero.

This is especially true, considering that the largest financial institutions in the world are already integrating these systems directly into the machinery of research, operations, onboarding, compliance, and decision-making itself. They are no longer content merely purchasing software infrastructure when they can instead purchase scalable cognition.

Clients Have Unprecedented Access

And if scalable cognition begins flowing through the financial industry the way spreadsheets, online trading, and index funds once did, advisors may eventually discover that the competition is no longer just the advisor across town. These days, it’s the intelligence layer sitting directly in the client’s pocket. Once clients gain access to tools capable of interrogating financial narratives, comparing strategies, reviewing fees, and synthesizing complex information in real time, the traditional informational asymmetry between advisor and client begins to compress.

A large language model can summarize a 90-page SEC filing in seconds, compare portfolio allocations, draft client reviews, explain tax strategies, synthesize research reports, identify inconsistencies across planning documents, translate dense financial language into plain English, and assist with coding, automation, and analysis. It is not always correct. Neither are people. The point is not perfection, but leverage.

Historically, firms scaled by adding people. Increasingly, firms may scale by increasing the cognitive throughput of existing people. If this still sounds abstract, translate it into the language the advisory industry understands best: smaller teams capable of producing more work, faster research cycles, more scalable client communication, and a gradual erosion of the competitive advantage that once came from simply being the person with access to the information.

Reactivity Misses the Point

So, what does the advisor do? First, stop self-soothing with defensive posturing. "People will always want a human relationship." "AI can't replace trust." "My clients aren't tech people." "AI hallucinates." None of those statements actually address the underlying shift.

The advisor who says "AI makes mistakes," while spending half the day manually re-entering account data into spreadsheets, toggling between PDFs, forwarding wholesaler decks, and relying on memory-based planning workflows is not making a serious argument. He is engaging in emotional risk management.

The time has instead come for offensive adaptation. However, by adaptation, I do not mean having the senior advisor's son build an Excel spreadsheet listing mutual fund expense ratios. Nor do I mean using AI as a glorified search engine. That’s like bolting a jet engine onto a horse carriage.

The real shift is architectural, and it requires advisors to start thinking in terms of systems, meaning structured workflows, knowledge repositories, project-based AI environments, context management, and automation layers. Each of those phrases describes something specific and buildable. They are not a vague aspiration toward "leveraging technology."

In plain English, it means spending less time hunting through emails, PDFs, spreadsheets, CRM notes, and wholesaler decks to reconstruct the same answers over and over again. Instead, it means more time building organized processes that make research, planning, communication, and decision-making faster, clearer, and more scalable.

Context Is Key

In a recent LinkedIn post, "Working Memory, Meet Wall Street," I wrote about one of the biggest misunderstandings advisors carry when using LLMs. They treat the model like a magic oracle while simultaneously overwhelming it with disorganized inputs.

An LLM has a context window that represents the amount of information it can hold in active working memory at once. Cram enough into it, and the signal-to-noise ratio degrades. That’s when the model starts optimizing for plausible-sounding rather than correct. This very real limitation is worth knowing and designing around.

However, this is not a uniquely “machine” problem. Your brain has a context window, too. You just call it your career. Twenty-plus years of market cycles, product narratives, client relationships, and institutional incentives have trained you on a very specific distribution of inputs. The difference is the LLM does not protect prior recommendations simply because admitting error would threaten its professional identity.

The Value of Clarity

The advisors who gain genuine leverage from these tools will not necessarily be the ones asking the cleverest prompts. They will be the ones building the cleanest cognitive systems around the models themselves. That means separating workflows into discrete projects, maintaining structured knowledge bases, using persistent reference documents instead of expecting the model to remember what you told it three conversations ago, and designing repeatable analytical frameworks where inputs, assumptions, and outputs are explicit rather than vibes-based. In other words, it means becoming less like product distributors and more like systems architects.

For example, instead of opening one giant chat window and bouncing from Roth conversions, to Social Security timing, to muni bond ladders, to a client’s concentrated Nvidia position, to an estate planning question all in the same conversation, advisors will increasingly need organized project-based workflows where information, assumptions, client details, and planning logic remain separated and structured. That may sound technical, but it really is not.

It can be as simple as maintaining separate projects for retirement income planning, tax analysis, portfolio construction, or business-owner strategies. At the same time, the advisor should be attaching clean reference documents, planning notes, investment philosophies, and repeatable frameworks. The goal is to make sure the model is operating from stable context rather than fragmented memory.

A Paradigm Shift

The moat is shifting. For years, the value proposition in this business was proximity to information. But what happens when the client sitting across from the advisor has their own LLM, capable of analyzing allocations, reviewing fee structures, comparing tax strategies, stress testing retirement assumptions, summarizing SEC filings, and interrogating financial narratives in real time? That’s where this is heading. Not theoretically or eventually, but right now.

Clients already have access to increasingly sophisticated frontier models through platforms like Anthropic’s Claude or OpenAI’s ChatGPT, as well as AI development environments like Replit. The latter allow ordinary users to generate analytical tools, calculators, dashboards, automations, and financial workflows from plain-English instructions alone.

A prompt into Replit can already generate a retirement calculator, portfolio analyzer, or budgeting application in minutes. A well-structured request to Claude can already produce comparative fund analysis, retirement income scenarios, or detailed critiques of investment proposals. Granted, the capabilities are still imperfect, but the direction of travel is becoming increasingly difficult to ignore.

Adapt or Fall Behind

These systems may not replace advisors overnight, and they may never fully replace the human dimensions of trust, judgment, and behavioral guidance. That’s not really the point.

Rather, the informational gap between institutions, advisors, and clients is narrowing rapidly. And industries built for decades around controlling access to information should probably pay very close attention when intelligence itself starts becoming broadly distributed.

That is ultimately what the Goldman Sachs–Anthropic partnership signals — not merely a faster chatbot or a more efficient workflow, but the beginning of a world where scalable intelligence becomes embedded directly into the operating system of finance itself. The advisors who recognize that shift early may adapt. The ones who dismiss it as another passing technology cycle may eventually discover that the industry already moved underneath them.

Matt Mattheisen is a Financial Advisor with Focus Financial and has more than two decades of experience in wealth management. Outside of advising, Matt is an author and programmer who has developed proprietary analytical tools focused on volatility, portfolio risk, and market behavior. He frequently writes and researches the intersection of financial services, emerging technologies, and artificial intelligence, with a particular interest in how AI is reshaping the future of advice.

---

A message from Advisor Perspectives and VettaFi: Discover something new! Click here to register for our upcoming webcasts.

More Financial Planning Topics >