AI Scientist v2 represents a new category of AI tooling — autonomous research agents that conduct the entire scientific discovery process end-to-end. Starting from a research topic or question, the system generates hypotheses, designs experimental protocols, writes the code needed to run experiments, executes those experiments, analyzes the results statistically, and produces a formatted manuscript ready for review. The agentic tree search approach means it can explore multiple research directions simultaneously and backtrack when a path proves unproductive.
The system builds on learnings from the original AI Scientist by Sakana AI, adding progressive exploration that manages compute budgets intelligently. Rather than committing fully to a single research direction, v2 allocates resources across promising branches and deepens investigation where results are most interesting. This mirrors how experienced researchers allocate their time — pursuing multiple leads while doubling down on the most promising findings.
With over 4,600 GitHub stars and an ICLR 2025 Workshop acceptance, AI Scientist v2 has gained attention from both ML researchers and developers interested in agentic system design patterns. The architecture patterns — tree search over complex task spaces, progressive resource allocation, and automated evaluation — are applicable beyond scientific research to any domain requiring systematic exploration. Sakana AI, backed by over $300M in funding, continues active development under a responsible AI research license.
