AI agents learn to steer their own research

A new architecture teaches AI research agents to decide what to investigate next, not just how, improving long-horizon autonomous research.