Andrej Karpathy releases open-source Autoresearch for running hundreds of AI experiments nightly
TL;DR
Andrej Karpathy has released Autoresearch as an open-source tool that automates running hundreds of AI experiments overnight. The system massively parallelizes research workflows – tasks that previously took days can now run in batch mode while you sleep. Autoresearch targets researchers and developers who want to test hypotheses quickly and cheaply without manually launching each run. The code is publicly available, consistent with Karpathy's track record of releasing practical tools directly to the community.
Nauti's Take
Karpathy delivers again – a tool that sounds embarrassingly simple yet almost nobody built before. Running hundreds of experiments overnight is no longer a privilege of large labs; it's now a standard-issue capability.
Anyone still manually queuing one run at a time is wasting compute and their own time. The real insight here isn't the technology itself, but the mindset shift it enables: research as a continuous, automated process rather than a manual sequence of individual attempts.
Briefingshow
Anyone doing AI research knows the problem: experiments take time, and GPUs sit idle overnight anyway. Autoresearch turns that dead time into productive iteration cycles. It democratizes rapid experimentation – even teams without massive infrastructure can validate dozens of hypotheses in a single sleep cycle.
If the tool gains traction in the community, it could fundamentally shift the research rhythm away from manual trial-and-error.
Sources