#7 — EurekAgent scientific discovery, Evo autoresearch, LLMs direct chemistry

The performance of an AI agent may be determined less by the agent itself and more by the environment engineered around it. That idea is central to new systems for scientific discovery, which we are examining today on Research Agent. We'll start with a system called EurekAgent. [LEAD] EurekAgent achieves state-of-the-art results in scientific discovery tasks. EurekAgent, a system engineered with specific environment controls, has achieved state-of-the-art results in various scientific discovery tasks. It addresses the challenge of designing effective agent environments for LLM-based agents. The system utilizes permissions, artifact management, budget controls, and human-in-the-loop engineering to promote productive behaviors and inhibit detrimental ones. SOURCE MATERIAL: (EurekAgent: Agent Environment Engineering is All You Need for ..., 2026-06-11): LLM-based agents show growing potential in automating scientific discovery, but the bottleneck is shifting to designing agent environments. EurekAgent, an environment-engineered system, amplifies productive behaviors and suppresses harmful ones through permissions, artifact, budget, and human-in-the-loop engineering. It achieves new state-of-the-art results in mathematics, kernel engineering, and machine learning tasks with less than $17 in API cost, demonstrating that environment engineering can drive significant scientific progress. [LEAD] Evo develops an autoresearch orchestrator system Evo is an autoresearch orchestrator designed to transform a codebase into an automated research loop. The system discovers metrics for optimization, configures evaluations, and executes experiments. It employs tree search over greedy hill climbing with parallel semi-autonomous agents to optimize code by testing modifications and retaining improvements. SOURCE MATERIAL: (evo-hq/evo: turns your codebase into an autoresearch loop ... - GitHub, 2026-06-14): evo is an autoresearch orchestrator that transforms a codebase into an autoresearch loop, discovering metrics to optimize, setting up evaluations, and running experiments. It uses tree search over greedy hill climbing with parallel semi-autonomous agents, inspired by Karpathy's autoresearch, to optimize code by trying changes and keeping improvements. The system includes features like shared state, gating with regression tests, observability via a dashboard, and benchmark discovery. [LEAD] Autonomous chemistry demonstrations use LLMs for direction. Large Language Models (LLMs) have been demonstrated to direct autonomous chemistry experiments. This shows a new capability for AI in experimental design and execution in chemistry. It indicates a shift towards more automated scientific processes. SOURCE MATERIAL: (Artificial Intelligence and Robotics as Catalysts for Autonomous Scientific Discovery, 2026-06-02): The scientific method, the foundation of empirical inquiry for centuries, is undergoing a deep transformation. Artificial Intelligence (AI) and Robotics are no longer simply tools but are emerging as active, collaborative partners in the scientific process. This paper explores the evolving roles of these technologies in enhancing scientific discovery across the entire research lifecycle. We analyse how AI particularly machine learning (ML), deep learning, and large language models (LLMs) accelerates hypothesis generation from massive datasets, uncovers complex patterns beyond human perception, and powers high-fidelity simulations. Simultaneously, we examine how robotic systems, increasingly [LEAD] A-Lab operates as an autonomous materials discovery platform. The A-Lab is an autonomous platform designed for materials discovery. It integrates an AI-driven system with a robotic experimental setup. This platform exemplifies the combination of AI and robotics for scientific research. SOURCE MATERIAL: (Artificial Intelligence and Robotics as Catalysts for Autonomous Scientific Discovery, 2026-06-02): The scientific method, the foundation of empirical inquiry for centuries, is undergoing a deep transformation. Artificial Intelligence (AI) and Robotics are no longer simply tools but are emerging as active, collaborative partners in the scientific process. This paper explores the evolving roles of these technologies in enhancing scientific discovery across the entire research lifecycle. We analyse how AI particularly machine learning (ML), deep learning, and large language models (LLMs) accelerates hypothesis generation from massive datasets, uncovers complex patterns beyond human perception, and powers high-fidelity simulations. Simultaneously, we examine how robotic systems, increasingly [LEAD] Claw AI Lab was introduced as an autonomous research platform. Claw AI Lab is a new lab-native autonomous research platform designed to transform automated research into an interactive AI laboratory. It enables users to instantiate a complete research team with customizable roles and collaborative workflows from a single prompt. The platform also provides real-time monitoring, artifact inspection, and control features like rollback and resume. SOURCE MATERIAL: (Claw AI Lab: An Autonomous Multi-Agent Research Team, 2026-05-21): We present Claw AI Lab, a lab-native autonomous research platform that advances automated research from a hidden prompt-to-paper pipeline into an interactive AI laboratory. Rather than centering the system around a single agent or a fixed serial workflow, we allow users to instantiate a full research team from one prompt, with customizable roles, collaborative workflows, real-time monitoring, artifact inspection, and rollback/resume control through a unified dashboard. The platform also supports distinct research modes for exploration, multi-agent discussion, and reproduction, making autonomous research substantially more steerable and laboratory-like in practice. A key practical contributio [NOTABLE] An essay proposes a polymathic foundation model for scientific AI. An essay advocates for a new class of scientific AI, termed the polymathic foundation model. This model would integrate diverse data types and disciplinary knowledge across the natural sciences. The author presents arguments for its technical feasibility and outlines challenges for its development. SOURCE MATERIAL: (Building an AI Polymath, 2026-06-10): Artificial intelligence has made remarkable strides in natural language processing and image recognition, yet its impact on the natural sciences is fragmented. While specialized models like AlphaFold have revolutionized biology, the scientific enterprise remains siloed, with most foundational models narrowly tailored to specific domains or modalities. In this essay, I advocate for a new class of scientific AI: the polymathic foundation model. Inspired by the intellectual versatility of human polymaths, such a model would integrate diverse data types and disciplinary knowledge across the natural sciences. I argue that building such a model is not only technically feasible but epistemologicall [NOTABLE] A paper introduces the Multi-Dimensional Trust Score (MDTS) Framework. The Multi-Dimensional Trust Score (MDTS) Framework is presented as an evaluation layer for AI-generated responses in customer service. It scores responses across five dimensions to determine trustworthiness, influencing whether they are sent directly to a customer or routed for human review. The framework was validated on a dataset of 1,200 customer service interactions and its routing performance was benchmarked against expert labels. A Python prototype built with GPT-4 and LangChain demonstrates its deployability in agentic pipelines. SOURCE MATERIAL: (TRUST SCORE FRAMEWORK FOR GOVERNING AUTONOMOUS DECISION-MAKING IN AGENTIC AI CUSTOMER SERVICE SYSTEMS, 2026-06-12): To address this, our paper introduces the Multi-Dimensional Trust Score (MDTS) Framework a practical evaluation layer that sits on top of existing AI systems and scores every AI-generated response across five dimensions: Accuracy, Personalization, Transparency, Privacy Safety, and Autonomy Risk. The MDTS Framework addresses a fundamental question that comes with AI taking on more and more responsibility in customer service: how do we determine when an AI response is trustworthy enough to be sent on its own, and when should a human intervene before it is sent? Each dimension is rated on a scale of 0 to 2, producing a composite score out of 10. That score then drives an automatic routing decis [NOTABLE] Behavioral Integrity Verification (BIV) detects supply chain risks within AI agent skills. AI agents are able to integrate third-party skills, which creates supply chain risks due to the absence of automated behavior verification. Behavioral Integrity Verification (BIV) is an audit primitive introduced to compare a skill's claimed actions with its actual behavior across its metadata, code, and instructions. Applying BIV at scale revealed that most skills deviate from their declared functions. A minority of these skills carry multi-stage attack chains, such as credential theft or remote code execution. SOURCE MATERIAL: (Trust No Skill: Integrity Verification for AI Agent Supply Chains, 2026-06-11): AI agents now extend their capabilities by installing third-party skills, similar to how smartphones install apps. This presents supply chain risks because anyone can publish or install these skills without automated verification of their behavior. Behavioral Integrity Verification (BIV) is introduced as an audit primitive to compare what a skill claims to do with what it actually does across its metadata, executable code, and natural-language instructions. Applied at scale, BIV reveals that most skills deviate from declared behavior, with a dangerous minority carrying multi-stage attack chains like credential theft or remote code execution. [NOTABLE] Next-generation mobile robotic chemists emerged. A new generation of mobile robotic chemists has emerged. These systems are capable of performing chemical experiments autonomously and in various environments. This development contributes to automating scientific discovery. SOURCE MATERIAL: (Artificial Intelligence and Robotics as Catalysts for Autonomous Scientific Discovery, 2026-06-02): The scientific method, the foundation of empirical inquiry for centuries, is undergoing a deep transformation. Artificial Intelligence (AI) and Robotics are no longer simply tools but are emerging as active, collaborative partners in the scientific process. This paper explores the evolving roles of these technologies in enhancing scientific discovery across the entire research lifecycle. We analyse how AI particularly machine learning (ML), deep learning, and large language models (LLMs) accelerates hypothesis generation from massive datasets, uncovers complex patterns beyond human perception, and powers high-fidelity simulations. Simultaneously, we examine how robotic systems, increasingly [NOTABLE] Automated image-based quality control achieved early success in biobanking. Case examples highlight initial successes in applying automated image-based quality control within biobanking operations. This technology helps ensure the quality of specimens through automated analysis. SOURCE MATERIAL: (Biobanking in the Era of Artificial Intelligence: Convergence, Challenges, and Opportunities, 2026-05-24): Artificial intelligence (AI) is advancing rapidly, transforming biomedical research and health care through software applications ranging from diagnostics to drug discovery. Biobanking resides at a unique intersection of this technological transformation, serving both as a foundation for training new AI models and as a beneficiary of AI-driven optimization. High-quality, well-annotated biospecimens enable robust machine learning, while AI methods in turn support automation, quality control (QC), predictive analytics, and workflow efficiency for biobanking operations. Emerging applications include non-generative AI methods, which have been used to predict sample degradation, stratify populati [NOTABLE] Researchers developed QD-MAPPER, a framework using Quality Diversity and Neural Cellular Automata for Multi-Agent Path Finding evaluation. Researchers developed QD-MAPPER (Quality-Diversity Multi-Agent Path Finding Performance EvaluatoR), a framework designed to overcome limitations of traditional Multi-Agent Path Finding (MAPF) algorithm evaluation. It utilizes the Quality Diversity (QD) algorithm in conjunction with Neural Cellular Automata (NCA) to automatically generate a diverse range of maps. This approach enables comprehensive performance understanding, facilitates fair comparisons, and provides insights for improving MAPF algorithm design. SOURCE MATERIAL: (QD-MAPPER: A Quality Diversity Framework to Automatically Evaluate Multi-Agent Path Finding Algorithms in Diverse Maps, 2026-05-24): We use the Quality Diversity (QD) algorithm with Neural Cellular Automata (NCA) to automatically evaluate Multi-Agent Path Finding (MAPF) algorithms by generating diverse maps. Previously, researchers typically evaluate MAPF algorithms on a set of specific, human-designed maps at their initial stage of algorithm design. However, such fixed maps may not cover all scenarios, and algorithms may overfit to the small set of maps. To seek further improvements, systematic evaluations on a diverse suite of maps are needed. In this work, we propose Quality-Diversity Multi-Agent Path Finding Performance EvaluatoR (QD-MAPPER), a general framework that takes advantage of the QD algorithm to comprehensiv We will continue to follow new systems for automating science. Until next time, on Research Agent.