Across a representative set of drafting tasks spanning NEPA document sections from 18 federal agencies, 19 experts found that generalized coding agents have the potential to speed up NEPA document drafting work by as much as 1 to 5 hours per subsection—up to roughly 15% reduction in drafting time—signaling a meaningful step forward in how AI can support complex government workflows.

Federal permitting is a complex and document-heavy process in government. Reviews often require reading hundreds of pages of technical reports, cross-checking information across multiple sources, and drafting detailed analyses that must meet regulatory requirements.

Through this collaboration, OpenAI and PNNL explored the power⁠(opens in a new window) of generalizing coding agents (in this case, Codex CLI) as an effective way to extract performance from reasoning models like GPT‑5 for research, technical analysis, and report writing tasks that involve a file system. By giving models access to a command-line interface (typically used for coding tasks), they can use more general strategies for solving a task than hand-crafted heuristics. These agents are required to:

For the United States to continue to grow its economy in this Intelligence Age⁠(opens in a new window), it must be able to build safely, responsibly, and quickly. As AI systems increasingly impact the physical world, we must understand their capabilities in domains like civil engineering, environmental, and regulatory analysis. Over time, advanced models will need to understand laws and regulations accurately as they help to invent new and safer technologies, protect natural resources, and meet human needs.