What is autoresearch?

An open-source AI agent system by Karpathy that autonomously runs and improves ML models. Humans define quantifiable goals, AI loops through experiments: modify code → run validation → keep or rollback → repeat.

How is autoresearch different from CI/CD?

CI/CD ensures code correctness; autoresearch explores if code can be better. It's the offensive player to CI/CD's goalkeeper.

What tasks can autoresearch handle?

Any quantifiable optimization: test coverage, API response time, compile time, memory usage, ML accuracy. Key requirement: a command-line metric.

autoresearch: Karpathy-Style Auto-Experiment Loop With Zero Real Code

Karpathy's autoresearch (8.6M views in 2 days) ported to Claude Code Skill (github.com/uditgoenka/autoresearch). Core design: ~630 lines of Python, Skill version is almost entirely Markdown. Workflow: human defines quantifiable metric → AI loops code changes → validate → keep/rollback → repeat. Achieved 11% speedup on already-optimized nanochat GPT-2 code after ~700 experiments. Applicable to any quantifiable optimization task.

autoresearch: Zero Lines of Real Code, Yet It Runs Experiments All Night

Andrej Karpathy's autoresearch project gained 8.6M views in two days. The core idea: humans define a quantifiable metric, AI loops through code changes overnight. The original is ~630 lines of Python; the Claude Code Skill version is almost entirely Markdown.

Karpathy demonstrated an 11% speedup on his already-optimized nanochat GPT-2 code after ~700 automated experiments. Applicable to any quantifiable optimization: test coverage, API response time, ML accuracy, compile time.

Key insight: the human role shifts from manual optimization to defining "what does better mean?" The AI handles the tactical execution of exploring the solution space.

In-Depth Analysis and Industry Outlook

From a broader perspective, this development reflects the accelerating trend of AI technology transitioning from laboratories to industrial applications. Industry analysts widely agree that 2026 will be a pivotal year for AI commercialization. On the technical front, large model inference efficiency continues to improve while deployment costs decline, enabling more SMEs to access advanced AI capabilities. On the market front, enterprise expectations for AI investment returns are shifting from long-term strategic value to short-term quantifiable gains.

However, the rapid proliferation of AI also brings new challenges: increasing complexity of data privacy protection, growing demands for AI decision transparency, and difficulties in cross-border AI governance coordination. Regulatory authorities across multiple countries are closely monitoring these developments, attempting to balance innovation promotion with risk prevention. For investors, identifying AI companies with truly sustainable competitive advantages has become increasingly critical as the market transitions from hype to value validation.

From a supply chain perspective, the upstream infrastructure layer is experiencing consolidation and restructuring, with leading companies expanding competitive barriers through vertical integration. The midstream platform layer sees a flourishing open-source ecosystem that lowers barriers to AI application development. The downstream application layer shows accelerating AI penetration across traditional industries including finance, healthcare, education, and manufacturing.

Additionally, talent competition has become a critical bottleneck for AI industry development. The global war for top AI researchers is intensifying, with governments worldwide introducing policies to attract AI talent. Industry-academia collaborative innovation models are being promoted globally, with the potential to accelerate the industrialization of AI technology.