Bridgewater CIO Greg Jensen says AI is already doing the equity analyst job

Greg Jensen, a chief investment officer at Bridgewater Associates, used a recent appearance on Bloomberg's Odd Lots podcast to argue that one of Wall Street's most familiar jobs is, in practice, already gone. His claim: AI systems now ingest both structured filings and unstructured company disclosures and produce forward estimates better than the humans who have historically done that work.

Jensen framed the comment as an assessment of where the technology sits today, not a prediction about where it is heading. He then offered a forecast in his own voice, saying the machine side of Bridgewater is producing outputs that stand next to the firm's long-running human process.

The substance of the claim is narrow and worth separating from the headline. Jensen is not saying every function performed by a sell-side or buy-side analyst is automated. He is saying the specific pipeline of turning filings, transcripts, and other disclosures into forward estimates is a task where the machine now beats the human on his own team. That is the airport-test version of what an equity analyst gets paid to do at the junior and mid levels.

The messenger matters as much as the message. Bridgewater's clients are pensions, sovereign wealth funds, and endowments, the buyers who have historically underwritten sell-side research indirectly through commission dollars and directly through allocations to funds that consume it. When the CIO of one of the largest hedge funds in the world tells that audience the human research layer has been matched by an in-house model, it is a signal to LPs about where the firm is spending its compute budget and where it is not spending its headcount budget.

The comment also lands on a live debate inside the research value chain. Consensus estimates, analyst ratings, and sell-side models are the scaffolding that a large share of the buy side still uses to frame a trade, even when they disagree with the number. If the marginal producer of that scaffolding is a machine at a client, the economics of who pays for the human version change.

The near-term signal to watch is language, not headcount. If more CIOs at large allocators begin describing their internal research stack as two parallel intelligences rather than one, the framing itself will do work on how the industry prices the human layer.