Ken Griffin, Citadel founder, says AI agents now complete complex finance tasks in hours or days that previously took master's- and PhD-level teams weeks or months

Citadel Founder Ken Griffin at the Stanford Leadership Forum 2026 discussing how AI agents compress elite analytical work from months into hours:

"It has been interesting to watch...work that we would usually do with people with master's and PhDs in finance over the course of weeks or months, being done by AI agents over the course of hours or days. So these are not mid-tier white-collar jobs. These are like extraordinarily high-skilled jobs being ...automated by agentic AI. And I got to tell you, I went home one Friday, actually fairly depressed by this because you could just see how this was going to have such a dramatic impact on society."

Full interview here:

https://youtu.be/Csjy_A3Kj9s?si=5hFZlp_L-nAs5aN3

@FundamentEdge Wow he used to be anti AI calling it all hype. What changed ???

@DrBeavisAI Agentic AI

https://www.fundamentedge.com/ai-accelerator

@FundamentEdge The ultimate irony of the knowledge economy is that the most expensive, highly specialized cognitive labor was actually the easiest to map out and automate.

@FundamentEdge We knew he would turn. Everyone will…just before everything makes a mid term top.

@FundamentEdge Claude code only really launched 1 year ago (sonnet 4.0). So we are now at the iPhone 3S stage. Imagine where we will be in 10 years.

@FundamentEdge Citadel can do this because their data was already structured and audited before agents entered the picture. Most enterprises are trying to do the same thing on top of 10 years of SharePoint and PDFs nobody can find. The capability gap is real. The precondition gap is bigger.

@FundamentEdge I recall being very surprised Griffin called AI for research hype last October. In my opinion there hasn't been a step change improvement in agentic capability since then; it seems like his perspective has just caught up to what was consensus almost a year ago.

@FundamentEdge It's guaranteed that he's deceived.

Plug in a reality-based definition of LLMs (e.g., including their stochastic nature) into the narrative and it collapses in absurdity.

@FundamentEdge this is the shift people are underestimating

agents don’t just make one person faster

they change how work should be assigned, checked, routed, and owned

the management problem moves from “who can do this analysis”

to “what workflow can safely absorb this much leverage”

@FundamentEdge Top-decile labor was the consumer base carrying services demand. Citadel automating PhD work in days means that base contracts. My K-shape recession piece assumed this hadn't started. Griffin says it has: https://crepesupreme.substack.com/p/whose-recession-is-it-anyway

So, you either believe the Lump of Labor theory and think that this puts everyone out of work...or that those same people can now be unleashed on new and different opportunities, expanding value at Citadel and the larger economy. Technological advances have always done the latter.

@FundamentEdge OpenAI and Anthropic are both very likely running their own hedge funds and consuming all the finance companies they have as customers insights. You can’t pay for lawyers to sue for data use if they put you out of business first.

@FundamentEdge If the multi-week, heavy-lifting research tasks are entirely automated by agents, how does the industry actually train its next generation of senior decision-makers?

@AndrewYang It is already happening. Our graduating students are seeing it first hand.

@FundamentEdge Mind you Citadel's entire modus operandi is based around research by the best and brightest. This isn't some mom and pops outfit that stumbled upon generative AI, we're talking about AI doing some heavy lifting in this scenario.

@FundamentEdge step change is doing a lot of work in that sentence. tooling got better, sure, but most of the productivity gain on our side came from the team trusting outputs less and putting cheap eval loops on everything the model touches. without those we just shipped faster and broke more.

@FundamentEdge AI does the analysis. Knowing what questions to ask, what the output misses, that's still the scarce resource. The toolkit changes. Judgment doesn't.