EvoScientist: Towards Multi-Agent Evolving AI Scientists for End-to-End Scientific Discovery

arXiv 2026 64.2 method

TLDR

EvoScientist is a multi-agent AI scientist framework with persistent memory and self-evolution for end-to-end scientific discovery.

Reasoning

The paper presents a novel evolving multi-agent system that addresses limitations of static pipelines by incorporating persistent memory and self-evolution. However, the abstract lacks explicit details on real-world experimental validation or benchmarks, making it unclear how the framework performs empirically.

Read-first score

Read-first score 64.2, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 83.

Recency 8%

100

Uses a gentle age decay so recent papers surface without erasing older foundations. 2026

Methodology quality 25%

Screens visible abstract and analysis fields for experiment, dataset, baseline, metric, and limitation evidence. markers=evaluation,experiment,result

Topical relevance 42%

69.2

Uses existing LLM keyword relevance scores normalized to 0-100. AI scientist,automated scientific discovery,autonomous research agent,automated research,literature review agent,survey generation,automated experimentation,experiment design agent,AI for scientific research,paper writing agent,research automation,scientific discovery agent

Reproducibility 25%

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=False; dataset=False; markers=code

Field roles

FrontierMethodology anchor

Rank sensitivity

Stability: volatile; rank range: 63.

Keyword Scores

AI scientist

automated scientific discovery

scientific discovery agent

autonomous research agent

automated research

automated experimentation

AI for scientific research

research automation

experiment design agent

literature review agent

survey generation

paper writing agent

Deep Analysis

Innovations

Evolving multi-agent AI scientist framework with persistent memory and self-evolution
Three specialized agents: Researcher Agent, Engineer Agent, and Evolution Manager Agent
Ideation memory that summarizes feasible research directions and records previously unsuccessful ones
Experimentation memory that captures effective data processing and model training strategies from code search trajectories and best-performing implementations

Methodology

EvoScientist employs three agents (Researcher, Engineer, Evolution Manager) and two persistent memory modules. The Evolution Manager distills insights from prior interactions into ideation and experimentation memories, which the Researcher and Engineer retrieve to improve idea generation and code execution. Evaluation compares against 7 open-source and commercial systems on idea quality (novelty, feasibility, relevance, clarity) via automatic and human evaluation, and code execution success rates.

Key Results

EvoScientist outperforms 7 state-of-the-art systems in scientific idea generation across novelty, feasibility, relevance, and clarity, and substantially improves code execution success rates through multi-agent evolution.

Resources

Paper