Jump Trading hands long-horizon quant research to GPT-6 Astra

Jump Trading's work with OpenAI is presented through Lucas Baker, who leads agentic research and development at the firm. His team builds the agents, harnesses and infrastructure that let quantitative researchers explore ideas in greater breadth and depth. The shift he describes is from AI as a tool for one-off code snippets and small bug fixes to a system that can develop entire codebases and services by itself, and can be steered more like a colleague: define a problem, an environment and evaluation criteria, then direct one or many agents in real time.

The article frames this against a broader progression, with 2024 agents writing a single file without mistakes, 2025 bringing entire codebases from scratch, and 2026 making progress on open research questions with many agents collaborating dynamically. Within Jump, Baker says the practical result is that agents can analyze findings, judge them against agreed criteria and redirect their own efforts rather than needing a person to review each round. That is what makes multi-day tasks pulling from many data sources feasible.

The stakes Baker describes hinge on control as much as capability. Because Jump operates in a heavily regulated industry where mistakes can carry financial and compliance consequences, the firm keeps human judgment at the center, with clear constraints and human acceptance at the end of the pipeline. Whether autoresearch becomes an ordinary part of a quantitative researcher's workflow may depend on whether those guardrails scale as fast as the autonomy they are meant to contain.