Jump Trading, a quantitative trading firm, has been using OpenAI's GPT-6 Astra to push AI agents from writing code snippets to conducting multi-day research projects. According to Lucas Baker, head of LLM R&D at Jump, the model's autonomy lets researchers define goals, environment, and evaluation criteria, then let agents work through complex workflows with minimal intervention.
The shift is described as moving from AI as a tool to AI as a colleague. Agents can analyze findings, judge them against agreed-upon criteria, and redirect their own efforts, enabling recursive improvement. Baker says this allows researchers to tackle longer, more ambiguous problems that require pulling from many data sources and interrelating everything into a comprehensive analysis.
Despite the autonomy, Jump keeps human judgment central. In a regulated industry, outputs like trading signals are scoped and reviewed like any other signal. Baker emphasizes the need for safe environments, clear constraints, observability, and human acceptance at the end. He envisions a future of "autoresearch" where fleets of agents coordinate with each other, but still start from human-defined systems.