The landscape of scientific research is undergoing a profound transformation. On March 28, 2026, SakanaAI released AI Scientist-v2, a system capable of autonomously generating hypotheses, running experiments, analyzing data, and producing peer-reviewable scientific manuscripts鈥攁ll without human intervention.
This isn’t merely an incremental improvement over its predecessor. AI Scientist-v2 represents a fundamental shift in how artificial intelligence approaches the scientific process itself.
From Template-Dependent to Truly Autonomous
The original AI Scientist relied heavily on human-authored templates, which limited its creative range to structured problems with known solutions. Version 2 discards these training wheels entirely.
The key innovation lies in the progressive agentic tree search guided by an experiment manager agent. This allows the system to explore research directions that wouldn’t fit into any predefined template.
The system autonomously generates research ideas, runs actual experiments using machine learning frameworks, analyzes the resulting data, and synthesizes everything into a coherent scientific paper.
How the System Works
AI Scientist-v2 operates through a sophisticated multi-stage pipeline. First, it uses a large language model to brainstorm and refine research directions, checking novelty against existing literature through Semantic Scholar integration. Then, it explores these hypotheses using a best-first tree search algorithm, running actual code experiments and debugging failures autonomously.
The experiment manager agent coordinates multiple parallel exploration paths, expanding research directions concurrently while managing computational resources efficiently.
Finally, the system synthesizes experimental results into a formatted manuscript, complete with literature reviews, methodology sections, and analysis of findings.
The First AI-Generated Peer-Reviewed Paper
Perhaps most remarkably, AI Scientist-v2 has already produced the first workshop paper written entirely by AI that was accepted through peer review. This milestone demonstrates that machine-generated scientific content can meet the basic standards of academic scrutiny.
Democratizing Scientific Research
The broader implications extend beyond any single research project. If AI systems can accelerate the pace of scientific discovery, the benefits could compound across millions of potential research questions that currently lack human investigators.
Scientists could use AI Scientist-v2 to explore preliminary hypotheses before committing significant time, or to automate the tedious process of running parameter sweeps and analyzing results.