How a simple spill separates autonomous buildings from automated ones
AutomatedBuildings.com's November Monday Live session defined building autonomy with a spill: automation sends an alert, while autonomy opens the work order and confirms the fix. That test helps facility teams judge platforms such as Honeywell Forge Cognition and PassiveLogic Level 3. The hard gaps sit in building data, leases and contractor handoffs.
This story was produced through MarketScale. See how Building Management teams put it to work with Customer Stories & Case Studies.
Key takeaways
A building counts as autonomous when it closes the whole loop on a problem: locate it, decide the response, open the work order, notify the right people and confirm resolution. An alert by itself is automation.
The earliest working examples of building autonomy started with something most facilities already own: the event calendar, used at the Perot Museum in Dallas and the Vancouver Convention Centre to pre-cool rooms before crowds arrived.
Read vendor pilot claims closely: the Voyager “alarm figure” is presented as a result, while Honeywell’s energy figure is framed as identified potential—keep that distinction when evaluating performance terms.
Free workspace
Turn your Building Management expertise into content.
Record interviews, organize footage, and write with AI on a free trial of the MarketScale platform for qualifying companies. No demo required, no credit card.
A spill on the floor is about as ordinary as building operations get. At AutomatedBuildings.com's November Monday Live session, it became the clearest working definition of an autonomous building anyone has offered this year.
Kelly Sinclair's Nov. 18 writeup of the discussion lays out the split. An automated building notices the spill and fires off an alert. An autonomous one works out where the spill is, decides what should happen, opens a work order, tells the right people and then checks that the problem is actually gone. Whether a robot holds the mop doesn't matter. What counts is whether the building can run the whole response across its systems and the people responsible for them.
The definition arrives at a useful moment. Autonomy is now a product claim: Honeywell Technologies introduced Forge Cognition on Oct. 14 with NVIDIA and Dell Technologies, and PassiveLogic sells what it calls Level 3 Autonomy. A facilities director hearing both pitches needs a test that gets past the label, and the spill supplies one. Where does the loop close, and what proves it closed?
Where the spill loop breaks
Detection is rarely the hard part. The panel pointed out that end-to-end workflows usually cross organizational lines, pulling in contractors, departments and service providers that today have little connection to one another. In the spill case, one system notices the problem and a different party has to confirm it is fixed. The loop closes only if information moves between them.
Context is the second gap. Sensor readings are of limited use if systems can't reliably say where a piece of equipment sits, what it serves and how it relates to the spaces and activities around it. And buildings don't hold still. Spaces get reconfigured, equipment gets replaced, occupants move. The panel argued that an autonomous system has to keep up with those changes, so the real job is keeping building information accurate for the building's whole operating life. Connecting the devices is the easy part.
Ken Sinclair, writing in AutomatedBuildings.com on Nov. 13, puts a face on the problem: data stuck inside proprietary building automation silos under point names like B1_FL2_VAV14_ZNT. His fix leaves working HVAC, lighting and electrical hardware in place. He adds an open data layer, such as Tridium Niagara, EdgeX Foundry or an MQTT broker, that translates BACnet, Modbus and LonWorks into accessible APIs, then tags every sensor and actuator using ASHRAE 223P, Brick Schema or Project Haystack.
A semantic model tells software what a data point is and what it connects to. In Ken Sinclair's example, once an AI agent knows Sensor_104 is a zone temperature sensor for Conference Room B served by AHU-2, it can reason about cause and effect without someone programming each relationship by hand.
The Coalition for Smarter Buildings is working on the same problem from the standards side. Its update to the panel covered how the open standards UDMI and LON might complement each other, which could extend the life of installed systems while adding new capabilities. For owners with large LON or BACnet estates, this is good news: the route to autonomy described here starts with tagging and translating the equipment they already have.
On the mechanical side, Ken Sinclair's framework comes closest to a closed loop through automated fault detection and diagnostics, or AFDD: continuous analytics that flag a leaking chilled-water valve, a stuck economizer damper or a drifting CO2 sensor before it wastes energy or breaks equipment. His self-healing sequences go further. When a non-critical sensor fails, the system switches to a backup, adjusts neighboring zone dampers to compensate and generates a diagnostic work order with root-cause context. For that class of fault, the spill test passes right up to the technician's handoff.
The event calendar as a control input
The most concrete examples from the session didn't need sophisticated AI. At the Perot Museum in Dallas and the Vancouver Convention Centre, event scheduling data was fed to building systems so they could match the activity expected. In one case, projected attendance let the systems pre-cool meeting rooms before large groups arrived. The building acted on what was about to happen instead of waiting for sensors to report what already had.
The earliest working examples of building autonomy started with something most facilities already own: the event calendar.
Convention centers show why. A single venue might host a trade show, a conference and a dinner for thousands, and each needs different ventilation, lighting, access, occupancy and room layout. Hotels face a similar mix, with guests, front desk staff, housekeeping and event organizers all expecting different things from the same building.
Ken Sinclair's framework turns this into a method built on intent signals: calendar platforms such as Google Workspace and Microsoft 365, access badges or anonymous occupancy sensors, used to predict room use hours ahead. Pair those with model predictive control, which combines a building's thermal model with weather forecasts, time-of-use utility tariffs and grid carbon intensity, and the building can pre-cool during off-peak hours or ride through a demand peak without anyone getting uncomfortable. For venues whose bookings already live in a scheduling system, this may be the easiest starting point. The data exists. The work is connecting it to the controls.
Forge Cognition and PassiveLogic against the test
Honeywell's Forge Cognition pairs real-time digital twins with AI agents that connect to existing customer deployments. Its listed applications are alarm triage, predictive HVAC asset health and energy simulation. Customers can run it in manual, semi-autonomous or autonomous mode, and operators can override it at any time. Pilots are running at NVIDIA's Voyager building in Santa Clara, the Fairmont Hotel in Dubai, NEX Mall in Singapore, Vanderbilt University and Honeywell's own campuses in Bangalore and Richardson, Texas.
Forge Cognition at NVIDIA's Voyager building
Honeywell Technologies
The verbs matter. The alarm figure describes a result. The energy figure is worded as potential the system identified, and a buyer should keep that distinction when writing performance terms into a contract. Alarm triage also maps neatly onto the first step of the spill test: deciding which events deserve a response at all. That triage runs through a True Alarm agent built on NVIDIA Nemotron open models, which classifies building-management alarms and improves through operator feedback, so operators' corrections shape what gets filtered out.
PassiveLogic comes at the problem through physics. Facilities Dive reporter Joe Burns, writing Nov. 10, described its Level 3 system, launched earlier this year and powered by NVIDIA, as a World Model of a building's windows, doors and mechanical systems that can change control settings on its own. Neda Sepasian, who leads AI and control systems design at PassiveLogic, said customers see labor savings alongside energy savings. As of that report, the company was targeting Level 4 later in 2026, which would add learning at the edge: systems coordinating resources among themselves and correcting their own models as equipment ages or conditions drift.
The company's July release framed Level 3 as control that is "no longer scripted," a trainable AI "white box" that stays "deterministic and traceable." Sepasian also pointed to smarter contracting and maintenance, with the system using the data it collects to simulate next steps, and called it "a use case that we see more and more." Traceability and contracting are where the spill loop's last step lives, so that use case would matter most if it reaches the contractor's side of the handoff.
The strongest objection to the spill test is that it sets the bar at the whole workflow, when much of the energy and labor value sits in the mechanical plant. Sepasian's labor point and the Voyager alarm figure both suggest the value shows up well before any building closes every loop. Both things can be true. A platform can do the heavy lifting on controls and diagnosis while the last steps, getting the right contractor moving and recording that the job is done, still depend on the organizational connections the panel described. Honeywell's own wording, coordinating actions across supported existing systems, sets the scope.
Leases, tenants and cloud outages
Some of the biggest obstacles aren't technical. The panel noted that lease agreements, ownership arrangements, security responsibilities and operating policies can be as limiting as the controls themselves. Even when software can reconfigure HVAC, lighting and access control, the people responsible may have no practical way to approve changes quickly. A building that hosts different activities through the day could be worth more than one built around a fixed schedule, but only if the paperwork allows it.
Each stakeholder also defines success differently. Occupants want comfort without juggling apps. Facility managers want visibility and fewer manual interventions. Property managers, asset managers and brokers need lease, cost, occupancy and financial data. Tenants increasingly bring their own enterprise platforms and expect the building to fit into them. Seen that way, interoperability is as much about coordinating organizations as connecting equipment.
Then there's failure. Recent cloud disruptions led the panel to argue that buildings must keep running when connected services go down, which favors edge control, local decision-making and designs with no single coordinating platform. Ken Sinclair spells out the guardrails: minimum airflow, freeze protection and pressure limits programmed at the controller, so the autonomous layer can optimize only inside safe limits. The vendor roadmaps seem to agree. Honeywell says Forge Cognition is expanding to far-edge deployments on NVIDIA DGX Spark, and PassiveLogic's Level 4 plan centers on learning at the edge.
One workflow, then the next
The panel's advice was to skip the building-wide overhaul. Let scheduling coordinate HVAC and lighting, or let a maintenance process link detection to corrective action, then prove the result and build on it. Each success improves interoperability and the data behind it.
Ken Sinclair's checklist puts that into practice. He suggests starting with high-value modules such as central chiller plant optimization, automated demand response through OpenADR or dynamic zone ventilation. Write machine-readable service levels between landlords, tenants and providers, such as holding 21°C to 23°C and under 800 ppm CO2, so agents have clear targets that can be audited. Secure the edge with BACnet/SC and TLS-encrypted MQTT, keep OT networks separate from IT, and run closed-loop control on a single air handler or chiller loop with a human approving changes before going fully unattended.
Before any pilot, write the spill test for one workflow: what event starts it, which system or contractor acts, and what record proves it finished. If nobody can name that last step, the pilot is an alarm upgrade.
The conversation continues at AHR Expo 2027 in Chicago, Jan. 25 to 27, where AutomatedBuildings.com runs free education sessions for the 27th year. On Jan. 27, David Holmberg and Michael Galler of NIST present Basic Cybersecurity for Building Owners. The thing to listen for in Chicago is how pilots are measured. Alert counts and modeled savings describe automation getting better. A closed work order, matched to the event that started it, is what a building that runs its own workflows actually looks like.
Sources
Your experts belong here
Every story in MarketScale Building Management starts with a company putting its facilities engineers, energy managers, and service technicians on the record. Buyers are already reading this topic. The only question is whose experts they find.
Owners and facilities teams pick on trust, and your engineers turn that trust into inbound conversations.
About the author
The MarketScale Newsroom reports on the companies, technologies, and trends shaping 16 B2B industries. It turns primary sources and expert commentary into clear, useful coverage for the people doing the work.