OpenAI teamed up with the Pacific Northwest National Laboratory to see if AI coding agents can slash the time it takes for federal permitting reviews. The collaboration created a benchmark called DraftNEPABench, which tested AI models on drafting environmental impact statements—a key part of the notoriously slow National Environmental Policy Act process. Nineteen experts across 18 federal agencies found that generalized coding agents could shave off 1 to 5 hours per subsection, cutting drafting time by up to 15%. That’s a big deal because these reviews often involve wading through hundreds of pages of technical reports and cross-checking data from multiple sources. The AI agents used Codex CLI, a command-line interface, to handle tasks like reading lengthy documents, verifying facts, and writing structured reports that meet strict legal requirements. Why it matters: This partnership shows how AI can tackle complex government workflows, potentially accelerating infrastructure projects from energy to transportation, which is crucial for economic growth in the Intelligence Age.