Idaho National Laboratory has launched a three-year, $60 million project to bring agentic AI to nuclear engineering, and the first year will go to finding out where AI agents actually help — and where they might hurt.

The project, called Prometheus, pairs INL with Argonne, Oak Ridge and Sandia National Laboratories, four universities and more than 20 industry partners. Its stated goal: compress the time it takes to get a nuclear power plant from concept to operation.

That timeline matters for R&D managers watching the nuclear portfolio. Nuclear plant development traditionally runs on decade-scale schedules, and any tool that shortens licensing, design or safety-analysis cycles would shift budget priorities across the sector. Whether agentic AI can deliver that compression is precisely what the first phase of work will test.

The principal investigator leading Prometheus said the team will spend year one evaluating where autonomous AI agents can be usefully inserted into nuclear engineering workflows — and where they introduce risk. That framing treats the technology as an open empirical question rather than a settled capability, an unusual stance in a field where vendor claims about agentic AI often run ahead of measured results.

For laboratory directors and research managers, the structure of the consortium is worth noting. Three additional national laboratories bring independent technical depth, four universities add academic scrutiny, and the roster of more than 20 industry partners signals that vendors and utilities expect deployable outcomes within the project's three-year horizon.

The open questions are concrete. Which parts of nuclear engineering — design iteration, regulatory documentation, safety analysis, operational support — can an AI agent handle without introducing errors that regulators cannot accept? Where does an agent's output need human verification so thorough that the time savings disappear? The project's first-year testing plan directly addresses those questions before scaling any tools.

No measured performance figures, benchmark results or specific deliverable dates have been released yet. The $60 million budget, the consortium membership and the three-year duration are the confirmed facts; claims about compressing plant development timelines remain projections the project itself aims to test.

The next milestone to watch is the outcome of the first-year evaluation, which will determine which agentic AI applications the consortium carries forward into the remaining two years of development.