Researchers at the U.S. Department of Energy's Argonne National Laboratory have introduced ChemGraph, an open-source artificial intelligence framework designed to automate and streamline computational chemistry and materials science workflows. Developed by computational scientist Murat Keçeli and postdoctoral fellow Thang Duc Pham, the system lowers technical barriers for researchers by translating plain-language scientific queries into complex, multi-step simulation pipelines. The framework was recently detailed in the journal Communications Chemistry. ChemGraph addresses a persistent bottleneck in materials science: the expertise required to execute atomically precise simulations. Traditionally, designing a virtual material and predicting its behavior demands extensive knowledge of quantum chemistry, software compatibility, input preparation, and iterative parameter tuning. ChemGraph resolves this through an agent-based automation architecture that delegates specific workflow components to specialized digital assistants. Each agent handles distinct tasks, including planning, execution, and data aggregation, allowing non-computational specialists to generate rigorous results without managing the underlying technical complexity. To optimize performance and cost, the development team moved beyond single-model approaches. Initial testing revealed that relying on a single large language model for all tasks led to inefficiencies and occasional failures as problem complexity increased. The team implemented a tiered strategy, utilizing larger reasoning models for high-level workflow planning and smaller, faster models for routine execution. The framework is hosted and scaled through the Argonne Leadership Computing Facility, leveraging the Aurora exascale supercomputer for computationally intensive quantum chemistry calculations and the ALCF Inference Service for secure, high-performance model access. This integration reduces external cloud dependencies and aligns with the DOE Genesis Mission to accelerate scientific discovery through AI. A critical design feature of ChemGraph is its commitment to scientific accuracy over generative speculation. Rather than relying on an LLM to fabricate answers, the system forces the model to execute physics-based simulations and return empirical data. This architecture minimizes hallucination risks and enables hypothesis testing for unstudied chemical phenomena. The open-source nature of the framework has already spurred external adaptations, including automated X-ray absorption near-edge structure simulations and high-throughput materials screening pipelines coordinated across exascale systems. Academic institutions are beginning to adopt ChemGraph as both a pedagogical resource and a research accelerant. By delegating workflow management to AI agents, researchers can redirect focus toward experimental design and scientific inquiry. Looking forward, Argonne researchers aim to operationalize ChemGraph as a chatbot-style service for the broader ALCF user community. The long-term objective is full autonomous discovery, where the system independently plans, executes, and refines complex computational campaigns with minimal human intervention. This evolution promises to compress development timelines for advanced batteries, efficient combustion systems, and novel materials, ultimately reducing costly laboratory trials and accelerating the translation of theoretical models into real-world applications.