October 6, 2026
AI agents spend days on research: how Jump Trading guides them
Jump Trading assigns GPT-6 Astra research spanning several days, while people set goals and regularly check progress.
At Jump Trading, AI agents gather data from multiple sources, test hypotheses and independently change the direction of their analysis.
The company uses ChatGPT for quantitative research: finding patterns in data and testing new hypotheses. According to OpenAI as of Oct 6, 2026, GPT-6 Astra also helps the team with everyday coding.
More frequent intervention before. Lucas Baker, head of language model research at Jump Trading, describes earlier agents as requiring frequent human intervention. With GPT-6 Astra, his team focuses more on setting goals, preparing the working environment and monitoring the work.
The researcher defines the task. They formulate the problem, provide a working environment and set criteria for evaluating the result. They then guide one or more agents as the work progresses. On tasks spanning several days, agents check their findings against the original criteria and independently choose the next direction for analysis.
Even on the longest tasks, researchers regularly check in with agents on data, runtime, priorities and the meaning of intermediate results.
Source