El Agente Gr\'afico: Structured Execution Graphs for Scientific Agents
arXiv:2602.17902v1 Announce Type: new Abstract: Large language models (LLMs) are increasingly used to automate scientific workflows, yet their integration with heterogeneous computational tools remains ad hoc and fragile. Current agentic approaches often rely on unstructured text to manage context and coordinate execution, generating often overwhelming volumes of information that may obscure decision provenance and hinder auditability. In this work, we present El Agente Gr\'afico, a single-agent framework that embeds LLM-driven decision-making within a type-safe execution environment and dynamic knowledge graphs for external persistence. Central to our approach is a structured abstraction of scientific concepts and an object-graph mapper that represents computational state as typed Python objects, stored either in memory or persisted in an external knowledge graph. This design enables context management through typed symbolic identifiers rather than raw text, thereby ensuring consiste
arXiv:2602.17902v1 Announce Type: new Abstract: Large language models (LLMs) are increasingly used to automate scientific workflows, yet their integration with heterogeneous computational tools remains ad hoc and fragile. Current agentic approaches often rely on unstructured text to manage context and coordinate execution, generating often overwhelming volumes of information that may obscure decision provenance and hinder auditability. In this work, we present El Agente Gr\'afico, a single-agent framework that embeds LLM-driven decision-making within a type-safe execution environment and dynamic knowledge graphs for external persistence. Central to our approach is a structured abstraction of scientific concepts and an object-graph mapper that represents computational state as typed Python objects, stored either in memory or persisted in an external knowledge graph. This design enables context management through typed symbolic identifiers rather than raw text, thereby ensuring consistency, supporting provenance tracking, and enabling efficient tool orchestration. We evaluate the system by developing an automated benchmarking framework across a suite of university-level quantum chemistry tasks previously evaluated on a multi-agent system, demonstrating that a single agent, when coupled to a reliable execution engine, can robustly perform complex, multi-step, and parallel computations. We further extend this paradigm to two other large classes of applications: conformer ensemble generation and metal-organic framework design, where knowledge graphs serve as both memory and reasoning substrates. Together, these results illustrate how abstraction and type safety can provide a scalable foundation for agentic scientific automation beyond prompt-centric designs.
Executive Summary
This article introduces El Agente Gráfico, a framework that leverages large language models and structured execution graphs to automate scientific workflows. The framework embeds decision-making within a type-safe environment, utilizing dynamic knowledge graphs for external persistence. This approach enables efficient context management, provenance tracking, and tool orchestration, demonstrating its potential in various applications, including quantum chemistry, conformer ensemble generation, and metal-organic framework design.
Key Points
- ▸ El Agente Gráfico framework integrates large language models with structured execution graphs
- ▸ Type-safe execution environment and dynamic knowledge graphs enable efficient context management
- ▸ The framework demonstrates robust performance in complex, multi-step, and parallel computations
Merits
Scalability
The framework's abstraction and type safety provide a scalable foundation for agentic scientific automation
Efficient Context Management
The use of typed symbolic identifiers enables efficient context management and provenance tracking
Demerits
Limited Domain Expertise
The framework's effectiveness may be limited by the large language model's domain expertise and training data
Expert Commentary
The El Agente Gráfico framework represents a significant advancement in scientific automation, offering a scalable and efficient solution for complex computations. The integration of large language models with structured execution graphs and type-safe environments can mitigate the limitations of traditional prompt-centric designs. However, further research is needed to address potential limitations, such as domain expertise and training data. The framework's implications extend beyond practical applications, as it can inform policy decisions on scientific automation infrastructure and contribute to the development of more explainable and transparent scientific automation systems.
Recommendations
- ✓ Further evaluation of the framework's performance in diverse scientific domains
- ✓ Investigation into the potential integration of El Agente Gráfico with other scientific automation tools and platforms
