Jump Trading's Head of LLM R&D Lucas Baker is deploying frontier models, specifically GPT-4o and what he references as next-generation systems like Astra, on tasks with no clean definition: multi-day analyses spanning numerous data sources and end-to-end service construction handled autonomously by agents.
The detail worth reading for is Baker's operational framing. This is not a pilot program or an experiment. He describes AI as embedded in the team's daily workflow, which means Jump, a firm that trades on millisecond advantages, has made a structural bet that long-horizon agents are reliable enough to touch core research and infrastructure work.
What comes next is the harder question the video raises without fully answering: how do you audit an agent that ran for three days across a dozen data sources and made consequential decisions at each step. Jump's answer to that, if they have one, is the reason to watch.
[WATCH ON YOUTUBE →]