Hey, I’m Matt. I’m a former Bloomberg News reporter, and you’re reading AI Street, where I report on how Wall Street uses AI. On Tuesdays, I highlight novel research, emerging use cases, and expert interviews.
Citadel Securities, Two Sigma, Bridgewater and other financial firms detailed AI use cases beyond basic summarization and document search at a conference in New York last week.
Examples included flagging internal projects at risk of running into trouble at Citadel Securities, proactively suggesting research questions at Bridgewater and automatically tracking who owns each AI agent and which tools it can access at Two Sigma.
Use Case Tracker
I’ve started a database tracking what financial firms have publicly said about how they’re using AI. So far, it includes 79 entries from 40+ companies, drawn from interviews, conference talks and technical posts, with links and key takeaways.
Paid subscribers can access the tracker by clicking the link below:
We’ve previously covered how firms are tailoring AI to their investment needs:
Below are five examples from the AI Engineer conference. More use cases are in the AI tracker.
PRESENTED BY BRICKROAD
The Best Dataset Isn’t For Sale — Yet
Months before Google agreed to pay $10 million for Spirit Airlines’ data, Brickroad had already flagged it.
The company’s AI agents scan news, earnings reports, app-store rankings and changes to terms of service for signs that a company is producing valuable data — and might be willing to license it. When a source looks promising, Brickroad contacts the company, secures a sample and helps the buyer evaluate the data before committing to a deal.
Citadel Securities Builds Internal ‘World Model’
Citadel Securities combines periodic project reports with Slack messages, meeting notes and code changes to track how projects are progressing.
In the talk, João Wenzel, chief operating officer for AI at Citadel Securities, said the team built up three years of historical records to test whether the system could flag troubled projects before their official status deteriorated.
In a backtest presented, 18 of 100 projects rated green and flagged by the system turned yellow or red within three months, versus 10 of 100 unflagged green projects.
Splitting the material among specialist agents, then having other agents review their findings, more than doubled useful project updates compared with a single-model pass.
Why are firms that have historically kept a low profile now more public?
A big reason why firms are now more public than they were, say, 10 years ago, (or maybe even five years ago) is the need to recruit AI/ machine learning talent. Frontier labs are fishing in the same PhD pool of math/ computer science/ engineering folks that were previously mostly sought after by quant firms.
Two Sigma Tracks Who Owns Which Agents
As the number of “AI employees” grows, firms need to know who owns them, what they do and which systems they can access.
John Feminella, senior vice president of architecture at Two Sigma, described automatically registering agents created through approved internal frameworks. The registry records each agent’s owner, purpose, tools and endpoints, making agents visible and reusable across the firm.
Feminella also argued that controls should reflect what an agent does. A tool summarizing support tickets should receive different scrutiny from one deploying production software. Feminella said he was speaking in a personal capacity.
His broader framework included restrictions on agent access, shared evaluations and checks on agents’ actions as they run.
Bridgewater’s AI Suggests What to Research Next
Bridgewater is building an investor workflow where AI proactively suggests questions worth investigating.
The system draws on Vault, which contains more than 100 million documents from 135 external sources, alongside internal research and discussions.
In the talk, Sam Green, team lead for applied AI at Bridgewater’s AIA Labs, described agents presenting investors with research topics, summaries and supporting evidence. Investors decide whether to dig deeper, and findings from their work with the agents are saved back into Vault for future research.
The repository also supports AI-generated macroeconomic forecasts and investment signals.
Apollo Uses AI to Prepare Loan Portfolios
At Apollo’s Redding Ridge Capital, preparing a proposed portfolio of 200 to 400 credits previously took two or three days, involving legal documents and Excel models.
Jack McDonald, a principal at Apollo and chief operating officer of Redding Ridge, said the firm now uses Rowspace to combine document libraries, trade and compliance systems, and current pricing to prepare portfolios in minutes.
McDonald said the process gets portfolios roughly 80% to 90% complete, leaving portfolio managers to apply their judgment and refine the result.
He also described automating reports on loan purchases, sales and settlements that previously occupied two or three employees for hours each week. Staff now check the output daily.
Coatue Maps Which Research Tasks AI Should Handle
Coatue’s head of AI, Frank Long, described breaking investment research into tasks best handled by software, machine learning, AI agents or people.
His example: processing Korean-language news about semiconductors and AI. That involves collecting, filtering and translating articles, then checking the information against the firm’s research and views.
Long said Coatue has preserved decades of notes on investment decisions and outcomes, providing internal context for AI systems.
The goal is better work at each step. A simple rule may beat a large model, while assessing portfolio implications may still require a person.
If you’ve read this far, I’d like to hear what you think of this format. Reply with “quick-hit” for this fast rundown of firms or “slow down” if you’d prefer a deeper dive into a particular use case.
Thanks for reading!
Love it. Exactly why we are excited to have you speak at "Accurate Insights 2026" (Weds Oct 28th)
All: Register here:
https://www.airmeet.com/e/935c31a0-b10d-11f1-ab73-f30d06bf1c71