# AI agents automate computational chemistry to accelerate catalyst discovery and industrial scale-up

> AI agents are compressing the 10- to 25-year catalyst discovery timeline by automating computational chemistry. Open Catalyst released 260 million calculations to train models.

- Type: AI news
- Page: https://completeaitraining.com/news/ai-agents-automate-computational-chemistry-to-accelerate/
- Published: 2026-06-16
- Updated: 2026-06-22
- Author: Jeroen Erne
- Jobs: Science and research
- Reading time: 2 min

AI agents are compressing the 10- to 25-year timeline for catalyst discovery and deployment by automating computational chemistry tasks and guiding experimental design. By integrating machine learning models with large-scale experimental datasets, researchers are moving from manual hypothesis testing to autonomous laboratory systems that can validate chemical reactions at industrial scale.

#### Accelerating molecular simulations

Varinia Bernales, a computational chemist who previously worked at Dow Chemical, spent hours manually creating files, submitting calculations, and troubleshooting. Now, the AI agent her team is developing, El Agente, handles these workflows. "With this tool, experimentalists will have another superpower," Bernales said. "We need it because we need to make progress faster and make quality of life better."

Researchers are replacing computationally demanding density functional theory simulations with **machine learning interatomic potentials**. Fernanda Duarte at the University of Oxford noted that these models can simulate complex mixtures and solvent interactions previously impossible to calculate. "We can do a calculation in 25 seconds that is perfect," Duarte said.

#### Building standardized datasets

The **Open Catalyst** project, a collaboration between Meta and Carnegie Mellon University, recently released results from over 260 million density functional theory calculations to train machine learning models. Developing these foundational computational modeling skills is a core focus of the [AI Learning Path for Research Scientists](https://completeaitraining.com/learningpath/ai-for-research-scientists/). Jahed Abed, a materials scientist at Google X, emphasized that open-source datasets lower the barrier to entry for external researchers.

Researchers are increasingly focusing on building standardized datasets to train machine learning models, a critical step for professionals advancing their work in [AI for Science & Research](https://completeaitraining.com/tag/science-and-research/). However, experimental data remains scarce because physical characterization is expensive and the chemical industry heavily protects its proprietary information. Abed argued that open sharing standards for catalyst formulation and processing data are necessary to solve commercially useful problems.

#### Overcoming scale-up hurdles

Moving a catalyst from discovery to industrial application remains the greatest bottleneck. Ted Sargent at Northwestern University noted that scale-up is often a trial-and-error process requiring extensive formulation. He proposed using video cameras and temperature sensors to record and monitor every material dispensed during large-scale synthesis.

"AI can help us discover hidden process variables, and thus to accelerate catalysis at scale," Sargent said. Bernales agreed that automation could reduce the empirical guesswork that often frustrates industrial replication efforts. Her team has already integrated machine vision into high-throughput metal-organic framework synthesis to improve data collection.

#### Why this matters for science and research professionals

Scientists and computational chemists must prioritize learning how to integrate AI agents into their existing laboratory workflows. Researchers who adopt tools that mine literature for structured data or automate density functional theory simulations will test hypotheses in days rather than weeks. Contributing to open-source datasets will also directly improve the accuracy of the machine learning models your team relies on for industrial scale-up.

---

This is the Markdown form of https://completeaitraining.com/news/ai-agents-automate-computational-chemistry-to-accelerate/. The whole site is mapped at https://completeaitraining.com/llms.txt.