autoresearch

autoresearch is an open-source project released by Andrej Karpathy in 2026 in which an AI agent repeatedly edits model-training code and runs experiments on its own.

1 articleLast mentioned

autoresearch is an open-source project published on GitHub in 2026 by AI researcher Andrej Karpathy. An AI coding agent modifies the training code of a small language model, runs training for a fixed amount of time, and keeps only the changes that improve a validation metric, repeating the cycle without human intervention.

Rather than editing the training code directly, the human writes the instructions given to the agent, while the agent designs and runs the experiments and judges the results. The name is also used more broadly for the practice of having an agent iterate on any task whose outcome can be scored by a metric.

This entry is based on AIPOST articles and widely known facts. If something is wrong, please send us a correction request.

Articles covering this entry

  • AI Agent Tuning: When to Rewrite Prompts and When to Retrain WeightsAI Performance ·

    In Andrej Karpathy's autoresearch approach, an agent edits a training script, runs it, keeps the change if the metric improves and reverts it if not.

    Read article

Found a mistake or missing an entry?

Wiki entries are based on AIPOST articles and widely known facts. If an entry is wrong, send us a correction request and we will check and fix it. Names from new articles are added to the wiki after the article is published.