ChemE-as-Code is an open‑source platform that lets AI agents built on large language models (LLMs) perform chemical‑engineering tasks while allowing engineers to review the results. The environment bundles calculation tools ranging from catalysis and process simulation to plant layout, HAZOP and life‑cycle assessment. Each tool is wrapped in a skill that tells a general‑purpose agent how to invoke it. Engineering files are stored in Git in a machine‑readable format, versioned, and the actual calculations run outside the LLM; every numeric value records its provenance. Automated checks report inconsistencies or stale results by file and line number.
In an illustrative ammonia case, the agents turned a textual task description into a preliminary basic‑engineering package—simulation files, P&IDs, layout, HAZOP, datasheets and relief design—for a design‑session API cost of only USD 224. By planting faults and engineering changes, the authors showed that the checks catch errors covered by their rules but miss those outside the rule set, so the checks guide human review rather than replace it. No independent engineer has yet reviewed the design.
Review: The platform demonstrates the feasibility of LLM‑driven agents in chemical‑engineering workflows, highlighting the importance of traceability and automated validation, yet it underscores that human oversight remains essential for safety and reliability.