ePlan: an AI-native workspace for US environmental compliance

NEPA work is slow, heavy on documents and depends on a few experts. Planners search for precedent projects, draft detailed regulatory documents and manage timelines with many phases. ePlan turns most of that into a guided conversation with an AI agent. My part was to design agent flows that planners can trust.

Environmental planners at the US Forest Service take a median of 105 days for a categorical exclusion and 392 days for an environmental assessment. Most of that time goes into finding similar past projects, writing the documents and coordinating timelines across teams and contractors. An AI that drafts regulatory documents has to be trustworthy, because a made-up precedent is worse than no draft.

• A median of 392 days per environmental assessment, in every office
• Regulatory writing where a wrong citation causes serious problems
• Multi-phase timelines shared between agencies and contractors
• Field work in places with no mobile signal

The core flow has three phases. First the chat interviews the planner. Then a research agent works in the background and finds precedent projects and the right Categorical Exclusions. After that, the workspace unlocks document generation. The planner can always see what the agent found and why, with the chat on the left and the project fields filling in on the right.

• A guided chat and a background research agent, side by side
• Task sheets that draft scoping letters, assessments and decision memos with citations attached
• A keyboard-friendly Gantt with separate owner and viewer states, plus shapefile-aware maps
• An offline-first mobile app that resolves edit conflicts when it syncs back

Every step of the agent's work can be audited. Each generated document links back to its source, the agent's progress is always visible, and nothing is submitted automatically. This rule shaped most of the design decisions, and it is the reason the product works in a regulated field.

One design system now covers desktop, mobile and the marketing site. It follows the structure of the agencies, with three levels (organization, office and project) and ten feature modules that each office can turn on or off. The platform is used by the US Forest Service, Tahoe National Forest and Tonto National Forest.

• A first compliant draft in one working session, compared with months of manual research
• Every AI-written passage links back to its source
• Field work continues offline and syncs when the connection is back