Artificial Intelligence

Citadel Securities says RAG can't track strategy

Citadel Securities built agentic memory to track strategy, splitting reported status from evidence. RAG alone called risk green.

StartupHub.ai Staff

By StartupHub.ai Staff, reviewed by Daniel Singer, Editor

2 min read

Joao Wenzel of Citadel Securities told AI Engineer his team built agents to keep a living model of the company itself. Over two thirds of staff write code, he said, and leadership asked if the same systems could help set strategy. His answer is an institutional memory loop that runs update, age, observe, grade and correct.

Citadel Securities says RAG can't track strategy - AI Engineer
Citadel Securities says RAG can't track strategy, AI Engineer

RAG is not enough.

Wenzel showed a March 16 status check where retrieval called a project green from slides and Slack threads and missed the real risk. His team keeps a three year longitudinal record, so it knew a renamed project was still Project Northstar from people and goals. It also splits reported status from evidence, flagging one green slide as yellow over an open dependency, and those flagged greens turned yellow or red at twice the rate in backtests.

Anyone who has parsed a BizOps deck knows why this breaks. Unpack the PowerPoint XML and you argue about the hex code for green. Screenshot it for a vision model and it squints at a pasted dashboard to tell a 31 from an 81. Wenzel said the fix was inelegant but effective, use both, then track continuity because Slack from five days ago will contradict the deck.

Every fact gets a half life in his design. Half the delivery dates changed within two months over three years, he said, and 90 percent slipped later than first promised. Specialist agents now break down the collateral instead of one pass over the corpus, which more than doubled insightful updates, and humans correct the record in Slack or email where stakes are high. Contrast that with Spotlight.ai, which describes a revenue team world model that plugs into Claude, Gong and the CRM via MCP for forecasting and stakeholder mapping.

The payoff at Citadel Securities is a weekly read that starts where deck and evidence disagree, plus a context layer other agents can reuse for clients. The hard parts left are entity resolution across four systems with no canonical ID and asking the right person the right question at the right time. Start collecting traces now or your agent will keep calling watermelons green.

Mentioned in this story

Sources

Get the free daily brief every morning

Every morning the day's AI news, new AI startups and new funding rounds. Free and takes a few minutes to read, unsubscribe at any time.

Join 64,000 other subscribers.

TopicsCitadel SecuritiesJoao WenzelAI EngineerProject NorthstarSpotlight.aiClaude

StartupHub.ai Staff

Written by

StartupHub.ai Staff

Editorial team

The staff writers of StartupHub.ai, ranging from investment analysts to avid AI tool users, early adopters and critical enthusiasts. Backgrounds span engineering, business and the arts. We hold every piece to rigorous standards of research and review.

Daniel Singer

Written by

Daniel Singer

Editor, reviewed this article

Daniel Singer is an AI expert who has been writing about, covering and building AI technology since 2018.

Test your Agent Readiness now

Paste your website. How well ChatGPT, Claude and the other agents can read and use a site, in about a minute.

Watch

The videos behind the news, one flip at a time

The feed mixes the videos from our articles with startup demos and you flip through them like TV channels for free with no signup.