LLNL Deploys AI Agents for Fusion Target Design on Supercomputers

Lawrence Livermore National Laboratory (LLNL) has introduced the Multi-Agent Design Assistant (MADA) system to accelerate inertial confinement fusion (ICF) target design. This innovative system integrates large language models with LLNL’s 3D multiphysics simulation code, MARBL, aiming to automate the creation of simulation decks. Researchers run MADA on the El Capitan supercomputer, one of the fastest globally with a peak performance of 2.79 exaFLOPs, as well as on its smaller sibling, Tuolumne. The system employs two AI agents: an “Inverse Design Agent” that converts hand-drawn capsule diagrams into thousands of simulation variants, and a “Job Management Agent” that schedules and manages simulation runs across high-performance computing (HPC) resources. The project, initiated in 2019 and led by LLNL physicist Jon Belof, began by exploring the unconventional combination of AI with shockwave physics. In a recent demonstration, an open-source large language model fine-tuned on MARBL documentation successfully transformed a hand-drawn capsule diagram and a natural language request into a complete simulation deck, running thousands of simulations to explore variations in ICF capsule geometry. Belof highlights that AI agents can drastically shorten design cycles, enabling researchers to evaluate hundreds or thousands of design concepts in parallel rather than just a few.

“Rather than the human running ensembles of simulations, they will be able to run ensembles of ideas.” — Jon Belof

Following LLNL’s December 2022 ignition milestone, the laboratory is now focused on developing a robust high-gain fusion platform for national security applications.

Two Agents Working in Concert

The MADA framework features two primary AI agents working collaboratively:

• Inverse Design Agent: Responsible for generating design concepts from hand-drawn inputs.
• Job Management Agent: Oversees the execution of large-scale simulation workflows, interfacing with the Flux scheduler and workflow tools like Merlin.
Giselle Fernandez, the Job Management Agent Team Lead, explains, “The Job Management Agent brings AI and HPC together to coordinate agents that handle resource management and workflow optimization at massive scales.” This approach has already yielded promising results. Using the Tuolumne supercomputer, the team has run tens of thousands of ICF simulations in ensemble studies. The simulation outputs train a machine learning model called PROFESSOR, which provides instant feedback to designers. According to Belof, “Once trained, the PROFESSOR model generates implosion time histories, radius as a function of time, that change instantaneously when the human designer changes the input geometry.” The project is funded by the National Nuclear Security Administration’s Advanced Simulation & Computing program. The MADA team includes LLNL researchers Charles Jekel, Rob Rieben, Will Schill, Meir Shachar, and Dane Sterbentz, alongside collaborators Nathan Brown from Sandia National Laboratories and Ismael Djibrilla Boureima from Los Alamos National Laboratory. Belof emphasizes the novelty of the system: “We are putting AI in the driver’s seat of a supercomputer, which is something that has never been done before.”

Federal Agencies Test Agentic AI with Mixed Results

LLNL’s advances come amid broader government experimentation with agentic AI systems. In June 2025, the FDA launched “Elsa,” an AI assistant trained on Anthropic’s Claude. Elsa received mixed feedback: some staff found it helpful for parsing test reports and reducing review times, while others noted its knowledge cutoff in April 2024 and occasional outdated responses. Elsa carries a disclaimer advising users to verify its answers due to potential errors. Meanwhile, NASA’s Goddard Space Flight Center is developing “text to structures” and “text to spaceship” projects, enabling scientists to describe spacecraft designs in natural language and receive AI-generated lightweight, efficient structures. Omar Hatamleh, Goddard’s chief AI officer, notes ongoing experiments with agentic workflows to identify bottlenecks in procurement and finance and automate routine tasks. A January 2025 Nextgov/FCW column predicted that AI agents will automate back-office tasks and optimize government workflows, freeing human staff for higher-value work. The article highlighted the U.S. Patent and Trademark Office’s AI-assisted search system as an example of AI improving efficiency and accuracy.

Industry Hype Meets Organizational Reality

Consultancies promote a grand vision for agentic AI. Capgemini’s July 2025 report “Rise of agentic AI” estimates an economic opportunity of up to $450 billion by 2028 from revenue growth and cost savings. However, the report found only 2% of organizations have deployed AI agents at scale, and trust in fully autonomous agents dropped from 43% to 27% within a year. It stresses prerequisites such as redesigning processes, reimagining business models, and balancing autonomy with human oversight. Similarly, McKinsey’s January 2025 report “Superagency” highlights that agentic AI can now autonomously converse with customers and handle follow-up steps like payments and fraud checks. While software vendors embed agentic capabilities into platforms to create “digital workforces,” only 1% of companies consider themselves mature in AI deployment.

---

Frequently Asked Questions (FAQ)

Understanding LLNL's AI Deployment for Fusion

Q: What is the MADA system and what is its purpose? A: The Multi-Agent Design Assistant (MADA) system is an AI-driven framework developed by LLNL to automate and accelerate the design process for inertial confinement fusion (ICF) targets. Its primary purpose is to streamline the creation of simulation decks for fusion experiments. Q: How does MADA utilize AI agents in fusion target design? A: MADA employs two main AI agents: an "Inverse Design Agent" that converts simple capsule diagrams into numerous simulation variations, and a "Job Management Agent" that handles the scheduling and execution of these simulations on high-performance computing resources. Q: What is the significance of deploying AI agents on supercomputers like El Capitan? A: Utilizing supercomputers allows MADA to run thousands of simulations in parallel, drastically reducing design cycles and enabling researchers to explore a much wider range of design concepts than previously possible. This leap in computational power, driven by AI agents, is key to advancing fusion research. Q: What are the potential benefits of using AI in fusion target design? A: AI agents like those in MADA can significantly shorten design cycles, allowing researchers to evaluate a vastly larger number of design variations. This parallel exploration of "ensembles of ideas" rather than just simulations can lead to faster innovation and more optimized target designs for fusion energy and national security applications. Q: What is LLNL's broader goal in fusion research after the recent ignition milestone? A: Following their December 2022 ignition milestone, LLNL is focused on developing a robust high-gain fusion platform for national security applications, and AI tools like MADA are crucial for achieving this objective.

---

Crypto Market AI's Take

The deployment of AI agents like MADA at institutions like Lawrence Livermore National Laboratory signifies a growing trend across various sectors: leveraging advanced AI for complex problem-solving and accelerated discovery. This mirrors the advancements we're seeing in the financial world, particularly in cryptocurrency. At Crypto Market AI, we understand the power of AI agents for navigating volatile markets, automating trading strategies, and providing insightful market analysis. Our platform utilizes sophisticated AI models to assist users in making informed decisions, much like how LLNL uses AI to optimize complex physics simulations. Just as LLNL aims to accelerate fusion research through AI, we aim to empower individuals and businesses by making sophisticated AI tools accessible for cryptocurrency trading and market intelligence, aiming to revolutionize finance through innovation.

More to Read:

• AI Agents Capabilities and Risks: The Growing Role of Artificial Intelligence
• Understanding AI-Driven Crypto Trading Tools

---

Source: Originally published at Research & Development World on August 4, 2025.