At the 2026 Energy Resources GIS Conference, David Nemeth of Energy Transfer showed software agents working across 140,000 miles of pipe in 44 states. The agents drew on GIS records, engineering documents and live regulatory sources, according to Esri's recap of the event. It was an impressive demo. It was also a bet that the records underneath are right.
That bet is about to be tested. Pipeline compliance consultancy RCP wrote that, effective January 1, 2027, all pipeline and LNG plant data submissions must align with major new spatial accuracy and attribute field updates. The scarce asset in pipeline geospatial work is not the AI tooling, which vendors are handing out in beta. It is clean, survey-grade data, and the regulatory calendar is about to price it.
The rulebook leaves no room for shortcuts
RCP's account of the changes is blunt. Every pipeline and LNG feature must be updated, it wrote, because of extensive revisions to the underlying data architecture. Operators cannot file "No Change Notifications" or use simple addition submission types for CY26 data. Revision code "N", meaning no change to the data feature, has been disabled for the 2027 filing period, and pipeline submissions stay barred from using it through 2028.
Omissions carry their own penalty. RCP wrote that they will be treated as deletions, forcing a systemwide rebuild and re-validation of active assets. Operators must also adapt their internal GIS to new Excel attribute templates and to the ESRI shapefile and geodatabase feature class templates distributed by PHMSA.
The deadlines are fixed. Gas transmission and LNG plant operators must submit finalized geospatial datasets by March 15, 2027, and hazardous liquid operators by June 15, 2027. RCP's advice was to start auditing current GIS databases against the updated schemas immediately. Nothing in that advice involves a language model.
The tools are ahead of the data
The software side is moving fast. In a blog post dated Nov 09, 2026, Esri said it is connecting ArcGIS to agentic ecosystems through MCP (Model Context Protocol). That equips agents such as Anthropic's Claude and Microsoft Copilot Studio to find places, plan routes and query spatial data. Esri's sample query asks an agent: "which 5 assets within 0.25 miles should I inspect next? Prioritize them based on the last inspection date". MCP support is in beta now, with general availability due in the next releases of ArcGIS Location Platform and ArcGIS Enterprise.
Esri sells the platform, so read its recap with that in mind. Even so, the vendor's own conference write-up carries a warning from the customer side. ExxonMobil's Lisa Zygo was direct that the data isn't ready yet for many of the AI tools on offer, and she advised keeping foundational QA/QC processes in place. Her advice on the tools themselves was short: "Don't just trust them." The oil and gas panel named data quality and governance as prerequisites for AI.
Esri's recap also shows where the work sits. Buckeye Partners and Geonamic Systems faced inline inspection data scattered across dozens of Excel spreadsheets from different vendors, and they built a centralized platform integrated with ArcGIS. Fugro's seafloor-mapping system detected more than 500,000 boulders from side-scan sonar data. It did so inside a governance framework with formal QA/QC validation at every step. Neither project was about a smarter model. Both were about data someone could stand behind.



