Morgan Stanley AI: What Its OpenAI Tools Change for Junior Staff

Morgan Stanley's AI is a family of named tools built on OpenAI's models, not a single chatbot. The AI @ Morgan Stanley Assistant, which in 2023 made Morgan Stanley the first major Wall Street firm to give staff a bespoke GPT-4 tool, answers financial advisors' questions from the firm's own research. Debrief sits in on client meetings and writes up the notes. AskResearchGPT lets investment bankers, traders and research staff query the firm's research library in plain English.

For someone starting at the firm, the change is where the hours go: less time finding a document, writing up a meeting or drafting the first email, and more time checking what the machine produced and owning the conversation with the client.

What Morgan Stanley's AI changes for a junior

Each tool takes a specific piece of junior work off the desk and leaves a checking job in its place. Mapped from what each tool is built to do, the swap looks like this by seat:

Where the support work goes

Note-taking is the clearest case. CNBC's launch report described Debrief as replacing note-taking that advisors or junior employees had been doing by hand, and a Menlo Park advisor put it from her side: "I don't have to rely on my team to jot down notes and action items anymore." In July 2025 the firm opened Debrief to the whole wealth field, not only advisors, so that the staff who support advisor teams could use it too.

Since 2024, staff on the institutional side have put their questions to the firm's research library instead of phoning or emailing the research department.

What stays with the person

The judgment call does not move. Katy Huberty, Morgan Stanley's global director of research, is often asked whether AI could ultimately replace the analysts who write the firm's research.

Her answer: "I don't see in the near future a path to just having the machine write the research report to generate the idea. I really think that it's humans who make the call and own the relationship, which is a really important part of the analyst job, or sales and trading job, or corporate banker job."

Why Morgan Stanley moved first with OpenAI

Those changes started in wealth management, where an advisor's job is to answer a client's question well and the answer usually sits somewhere in the firm's own research.

Morgan Stanley began working with OpenAI in 2022. When it announced the partnership in March 2023, its wealth division was one of a handful of launch organizations for GPT-4 and, in the firm's words, "currently the only strategic client in wealth management" with early access to OpenAI's new products. The goal was to put the firm's research within instant reach of roughly 16,000 advisors.

Why OpenAI, and why then

Jeff McMillan, who ran analytics, data and innovation for the wealth business, led the work. In March 2024 he became the firm's first head of firmwide AI, and he has since left the firm. On the choice of partner, he said: "We comb through hundreds of startups and tech firms to find technologies that can help enrich and improve the Financial Advisor and client experience, while aligning with appropriate controls, and OpenAI stands out."

The payoff he promised was "like having our Chief Investment Strategist, Chief Global Economist, and Global Equities Strategist on call for every Financial Advisor 24/7." It was not the firm's first AI: since 2018 its wealth business had used machine-learning algorithms that prompt advisors to contact clients, alongside an in-house engine called Next Best Action that tailors messages to clients and prospects.

What was new was the speed. "I've never seen anything like this in my career, and I've been doing artificial intelligence for 20 years," McMillan said in September 2023. "We saw a window of opportunity that was just completely disruptive, and I think as an organization, we didn't want to get left behind."

What is the AI @ Morgan Stanley Assistant?

The AI @ Morgan Stanley Assistant is a chat window for financial advisors and their support staff that answers questions about markets, recommendations and internal processes from about 100,000 research reports and documents the firm has vetted. Advisors had to learn to ask it in full sentences, as they would ask a colleague, rather than typing keywords into a search box.

Its reach grew quickly. David Wu, then head of firmwide AI product and architecture strategy, described the jump: "We went from being able to answer 7,000 questions to a place where we can now effectively answer any question from a corpus of 100,000 documents."

Adoption followed fast, and the firm put a number on each stage:

300

Advisors in the first pilot

March 2023

98%

Advisor teams using the Assistant

June 2024, per Morgan Stanley

July 2025

Opened to wider wealth staff

beyond advisors, per the firm

How does Debrief work?

Debrief takes over the meeting notes described above. Launched in June 2024, it joins an advisor's Zoom call with the client's consent, which has to be given every time. It records the meeting, surfaces action items, summarizes the key points, drafts a follow-up email for the advisor to edit and send, and saves a note into Salesforce. It also runs on mobile, for meetings outside the office.

Morgan Stanley planned to put it in front of roughly 15,000 advisors within weeks of launch, in a wealth business that hosts about a million Zoom calls a year. "The truth is, this does a better job of taking notes than the average human," McMillan told CNBC at the launch.

Kaitlin Elliott, whose team rolled the tool out, reported what advisors noticed: "They're more engaged with clients, and follow-ups that used to take days now happen within hours." David Wu has asked why a tool built for advisors speaking to clients could not also serve "the investment banker speaking to the CFO."

What half an hour a meeting adds up to

In June 2024 Ted Pick told investors that AI could save the firm's financial advisers between 10 and 15 hours a week, pointing to the note-taking tool, and called it "potentially really game-changing." One advisor's own numbers show how a figure like that is built:

1. Don Whitehead, a Houston advisor who tested Debrief, said it saved him "about half an hour per meeting" on note-taking.
2. He also said he runs "four, five or six meetings a day." At half an hour each, that is two to three hours a day.
3. Over a five-day week, two to three hours a day comes to 10 to 15 hours, the same range Pick gave.

What is AskResearchGPT?

AskResearchGPT is Morgan Stanley's generative AI research tool for the institutional side of the firm: investment banking, sales and trading, and research. Announced in October 2024 after a rollout that began that summer, it lets staff look for data, pull out insights and summarize information from the firm's body of research, which grows by more than 70,000 proprietary reports a year. It was the first generative AI tool built for that side of the firm.

It is a generative upgrade of AskResearch, an earlier chatbot that could already pinpoint data, key findings and analyst ratings. Drawing on multiple research products at once, the new version can take on more complex questions, and Morgan Stanley says staff ask it three times as many questions as they asked the older, traditional-AI tool in use since 2017. It sits in the browser, in Microsoft Teams and in Outlook.

The payoff shows most among salespeople who field questions from hedge funds and other institutional investors. "We found that it takes a salesperson one-tenth of the time to respond to the average client inquiry," Huberty said at launch, calling the tool "a game changer from a productivity standpoint."

A worked example: answering a client's question

Put the documented features together and a typical request runs like this:

1. A hedge fund client asks a salesperson for the firm's view on a sector.
2. The salesperson asks AskResearchGPT in a full sentence and gets an answer drawn from several reports, each one linked.
3. A workflow Morgan Stanley patented turns that answer into an email draft in one click.
4. Before sending, the salesperson opens each linked report and checks the rating, the figures and the publication date against the latest note.
5. They rewrite the draft in their own voice and send it.

Steps two and three are where the time saving comes from. Step four is the checking job from the seat table, and it is the part a junior is judged on.

The rest of the toolkit: code and tax advice

Two less publicized tools show how far the approach reaches, one for the firm's engineers and one for its advisors.

DevGen.AI

Morgan Stanley launched DevGen.AI in January 2025. Built on GPT models and trained on the languages in the firm's own code base, including ones customized for the firm, it translates old code into plain-English specifications, and developers then do the rewriting in modern languages such as Python. In its first five months it worked through nine million lines of code and saved the firm's 15,000 developers an estimated 280,000 hours.

Mike Pizzi, then the firm's global head of technology and operations, gave the case for building rather than buying: "We found that building it ourselves gave us certain capabilities that we're not really seeing in some of the commercial products." He added, "We saw the opportunity to get the jump early," and said the firm would not reduce its software engineering workforce because of the tool.

The Roth Conversion Analyst

In February 2026, Jed Finn, who runs the wealth business, pointed to an AI tool called the Roth Conversion Analyst. It pulls in a client's data, takes the advisor's assumptions about the future and recommends whether the client should convert a retirement account.

How does Morgan Stanley test its AI?

In February 2023, a month before the firm announced its first tool, Morgan Stanley analysts wrote that ChatGPT occasionally "hallucinates and can generate answers that are seemingly convincing, but are actually wrong." The tools the firm went on to build answer only from its own vetted content, and every AI use case is tested against real-world use cases, with expert feedback, before it ships.

Before the first tool launched, the firm spent months curating documents and having experts test the answers. The testing framework, known as evals, then grew in stages:

• Summarization evals came first. Advisors and prompt engineers graded the model's summaries of research and process documents for accuracy and coherence, and the team refined its prompts until the output held up.
• Translation evals followed, for clients who work in other languages, along with work to fine-tune how the tools retrieve documents as the library grew.
• The meeting tool got its own test sets, covering different kinds of meeting and checking that the model caught every critical action item without introducing errors.
• A daily regression suite of sample questions was added to find weaknesses in the live systems.

For a candidate, evals are the most transferable idea in Morgan Stanley's approach. The whole method depends on people who know what a good answer looks like and can grade a machine against it.

Which AI models does Morgan Stanley use?

Morgan Stanley has run its named tools on OpenAI's models from the start, and in a March 2024 memo its co-presidents called OpenAI the firm's "exclusive partner."

One partner, worked closely

Elliott described the working loop: "Based on all the questions we input and outputs we're getting, we'd sit with OpenAI and say, 'What can we change about our retrieval methods to help the accuracy we need at Morgan Stanley?'" Wu said one of the first questions the team hears is whether OpenAI will use the firm's information to train the public ChatGPT, and he called OpenAI's willingness to ensure zero data retention "really impactful."

In September 2026, OpenAI named Morgan Stanley a design partner on ChatGPT for Financial Services, a product for banking and research work that is covered in the analysis of whether AI will replace investment banking analysts.

JPMorgan went the other way, building one portal onto several outside model providers so that it would never depend on a single one.

The Anthropic exception

The Anthropic line is narrow. In April 2026, Pick told analysts the firm was "permissioned on" Claude Mythos Preview, which Anthropic released only in limited form, for defensive security work, because hackers could exploit what it can do to software. He raised it in a discussion of cyber resilience, not advisor tools.

What Morgan Stanley's leaders have said about AI and jobs

The firm's wealth leaders have said since 2023 that the advisor's relationship with the client stays human; its technologists talk more about what the work itself becomes.

Ted Pick's view

Chief executive Ted Pick frames AI as a tool for the people the firm already has. His 2026 letter to shareholders says "Morgan Stanley was an early adopter of AI" and calls the tools "a productivity enabler" that "augment human judgment and advice."

On the April 2026 earnings call he told analysts, "AI is our friend, okay?" He described a move away from "pure efficiency exercises," such as replacing what "might have been a call center or what might have been an operational function," toward "something that over time becomes a productivity phenomenon." That includes "super agents" for the wealth business and a client agent inside the equities electronic trading platform.

His July 2026 remark that AI is one of the year's two defining themes sits in that same analysis of analyst jobs.

What has happened to headcount

In March 2026 the firm cut about 2,500 roles, roughly 3% of its workforce, across investment banking and trading, wealth management and investment management; financial advisors were not affected. The reasons it gave were business priorities, location strategy and individual performance, and it said it plans to add resources in other areas. No Morgan Stanley statement ties AI to the size of its analyst class.

What a candidate should know about Morgan Stanley's AI

Three things are worth carrying into an interview at Morgan Stanley, whichever desk it is for:

1. Know why the firm built its own tools. Answers come from vetted firm content with links to sources, the model provider keeps none of the data, and each use case is tested before launch. That is the answer to "why not just use a public chatbot?"
2. Talk about the tools as an outsider. They run inside the firm, so discuss them through what Morgan Stanley has said publicly, with dates, rather than claiming hands-on use.
3. Have a view on the relationship argument. From Saperstein in 2023 to Finn in 2026, the firm's wealth leaders say the relationship stays human. Be ready to say which judgment calls stay yours when the draft arrives in seconds.

A structured way to rehearse those answers is the interview readiness tool, and the guides cover what other banks have built for themselves.

The bottom line

Morgan Stanley's AI is not one chatbot but a set of tools, each built for one kind of work: an answer engine for advisors, a note-taker for their meetings, a research assistant for bankers and traders, a translator for old code. The firm moved first, and it built those tools to answer only from its own vetted research, weeks after its own analysts had warned that ChatGPT could sound convincing and still be wrong.

What the tools change for a junior is the shape of the day. Less of it goes on finding and writing up; more of it goes on checking the draft and owning the client, the part the firm's wealth and research leaders say stays human.