OpenAI has published a timeline for autonomous research agents. By September , they plan to ship an "autonomous AI research intern" that can take on specific research problems independently. By , they plan a fully automated multi-agent research system.
Their strategy rests on two bets: continued algorithmic improvement and scaling test-time compute. For hard problems, they envision AI systems using entire data centers to think for weeks.
AutoResearch is already a working version of the target, just scoped to ML training optimization. The gap between Karpathy's -line script and OpenAI's ambition is the breadth of problems the system can handle, not the core loop.