Honeywell Forge Cognition Launches With Autonomy Disabled

Honeywell Technologies launched Forge Cognition on Wednesday, an artificial intelligence platform designed to manage the heating, cooling, security and power systems inside commercial buildings. Despite engineering the software to operate fully autonomously, the company has deliberately locked that capability behind a manual override.

Chief Technology Officer Suresh Venkatarayalu confirmed the platform will run exclusively in manual or semi-autonomous modes for at least two to three years. Operators will review every decision made by the AI agents before any physical equipment responds. The restriction applies even though the underlying architecture is capable of coordinating actions across building infrastructure without human intervention.

The decision reflects a broader hesitation within industrial automation regarding how much authority to grant generative AI models operating in physical environments. Unlike software bugs in enterprise applications, errors in building management systems can trigger fire suppression failures, lock occupants out of secure areas or cause catastrophic energy waste. Honeywell’s leadership determined that operators must build trust through continuous oversight before ceding control, setting a self-imposed standard of flawless accuracy as the prerequisite for unlocking full autonomy.

Forge Cognition represents the first major product release from Honeywell Technologies following its corporate breakup. After activist investor Elliott Management pressured the conglomerate to restructure, Honeywell spun off its advanced-materials division into Solstice Advanced Materials in October 2025 and separated its aerospace business into an independent publicly traded entity in June 2026. The remaining company now focuses entirely on automation for buildings, factories and process plants, serving an installed base of 350,000 sites globally.

Rather than launching the platform with theoretical demonstrations, Honeywell secured Nvidia as its first deployment site. The arrangement originated during the CES trade show in January when Nvidia executives challenged Honeywell CEO Vimal Kapur to identify a real-world facility willing to test the technology. Kapur countered by asking Nvidia to serve as the proving ground.

Within 90 days, Honeywell engineers deployed the system at Voyager, Nvidia’s 750,000-square-foot headquarters in Santa Clara, California. The teams monitored approximately 900 pieces of equipment against five specific performance targets ahead of Honeywell’s investor day in June.

The most immediate operational impact involved alarm fatigue. Commercial building management systems are historically configured to trigger alerts for any deviation from baseline parameters, generating massive volumes of false positives. At Voyager, between 90% and 95% of security and access-control alarms required no action, yet consumed the attention of eight or nine personnel per shift. Honeywell trained Nvidia’s open-source Nemotron language model using its own proprietary engineering data to filter critical alerts from background noise, ultimately reducing unnecessary Building Management System notifications by up to 90%.

Energy analysis revealed different inefficiencies. The AI agents mapped utility consumption by floor, zone and time of day, identifying heating systems running at a static 58-degree baseline regardless of external weather conditions. Honeywell and Nvidia calculated that correcting these thermal setpoints could yield nearly 10% in energy savings, though those reductions remain projected rather than realized.

Technically, the platform layers AI agents and digital twins over legacy control systems. Nvidia’s Omniverse simulation software generates the virtual replicas of physical spaces, while Dell provides the on-premises servers required to process the data locally. This edge computing approach addresses latency concerns inherent in cloud-dependent architectures, ensuring that commands affecting physical machinery do not rely on external network stability.

Traditional building automation relies on deterministic logic: sensors detect a threshold breach, and pre-programmed rules execute a response. Forge Cognition introduces probabilistic reasoning, allowing the system to interpret sensor data contextually and adapt based on operator feedback. The objectives also vary drastically by client. While Honeywell prioritized energy conservation during testing at its Bangalore campus, Nvidia explicitly instructed the engineering team to prioritize occupant comfort at Voyager.

The platform arrives amid a severe shortage of skilled facilities engineers. Industrial replacement parts frequently carry lead times of six to 18 months, making predictive maintenance essential for avoiding prolonged downtime. Nvidia’s head of operations and real estate, Scott Ekman, noted that the AI currently performs comparably to his most experienced building operators—expertise that took over a decade to cultivate and cannot be scaled fast enough to support Nvidia’s expanding real estate footprint. To address this, Nvidia mandated that Honeywell’s forward-deployed engineers train its internal staff on providing continuous feedback to the models before finalizing a commercial service contract.

“Physical AI is bringing new intelligence to the complex systems that run our buildings and industrial facilities,” said Deepu Talla, Nvidia’s vice president of robotics and edge AI.

Following the August milestone at Voyager, the relationship is transitioning from a pilot program to a commercial deployment. Nvidia is currently evaluating two or three additional U.S. facilities and one location in Berlin for integration. Globally, Honeywell is piloting the technology at the Fairmont Hotel in Dubai, NEX Mall in Singapore, and its own campuses in Richardson, Texas, alongside Bangalore. The company has selected 12 test sites across 12 distinct industries for the next six months, including a plant operation in Nigeria, requiring the models to be retrained for entirely new operational parameters at each location.