Honeywell launches Forge Cognition, AI agents on a live digital twin to run buildings and campuses

Honeywell Technologies launched Forge Cognition on 14 October 2026 in Charlotte, North Carolina: a self-contained agentic AI and simulation platform for operating buildings, campuses and industrial sites. It pairs real-time digital twins with AI agents that assess conditions, simulate operating scenarios and coordinate actions across the building systems a facility already has. NVIDIA supplies the AI models and simulation libraries; Dell Technologies supplies the on-site AI infrastructure. Honeywell says its installed base exceeds 300,000 sites.

What the platform does

Honeywell names three first applications. Alarm management agents classify and verify building management system alarms to cut false positives. Predictive asset health agents read live HVAC (heating, ventilation and air conditioning) data to flag emerging equipment faults and guide maintenance. Energy intelligence uses digital twin simulations to test operating changes that reduce energy use while keeping occupants comfortable. The platform connects to existing deployments and uses a structured knowledge base, built on Honeywell's engineering expertise, to map how equipment and systems relate.

NVIDIA Nemotron open models power the agents, including a True Alarm agent that improves through operator feedback. Omniverse libraries provide physically accurate simulation, and Cosmos models with the Metropolis VSS Blueprint let vision agents search and summarise live or recorded sensor data. Dell provides PowerEdge XE servers, data platform and services through the Dell AI Factory with NVIDIA.

Pilot results and the autonomy question

At NVIDIA's Voyager campus in Santa Clara, California, which Fortune describes as roughly 750,000 square feet with about 900 pieces of equipment, Honeywell reports up to a 90% average reduction in unnecessary alerts, up to 10% potential reduction in energy and utility use, and a comfort score above 95%, a weighted average of temperature, humidity and CO2 across zones. Fortune notes that the energy figure is a projection, not an achieved saving. Other pilots run at the Fairmont Hotel in Dubai, NEX Mall in Singapore, Vanderbilt University and Honeywell's own campuses in Bangalore and Richardson, Texas.

The platform offers manual, semi-autonomous and autonomous modes, and operators can override it at any time. Chief Technology Officer Suresh Venkatarayalu told Fortune: "We have not opened up an autonomy right now." He expects it to run with people in charge for probably the next two to three years, with customers deciding when to switch. Customers can now contact Honeywell to scope deployments; pricing was not disclosed.

Why it matters for delivery teams

An operational twin of this kind is only as good as the asset data it receives at handover: equipment lists, system relationships, sensor points and consistent naming. Teams producing as-built models and asset registers for owners of large campuses should expect more requests for clean, structured equipment data that an AI agent can map without manual rework, and should agree that data format with the operator before commissioning.

Image: The main entrance to NVIDIA's headquarters at 2788-2888 San Tomas Expressway, Santa Clara, California, in 2022. Photo: Coolcaesar, CC BY-SA 4.0, via Wikimedia Commons.