Johnson Controls unveils new era of building optimisation with agentic AI

Johnson Controls is introducing a range of new features on its OpenBlue platform using ‘agentic’ AI to learn a building’s operations; to optimise performance; and to automate critical functions.

The manufacturer believes there is huge scale for energy saving and productivity gains, having seen early versions of the platform producing significant savings from trials.

At the company’s recent conference for buildings operators, Dr Irina Koitz, the company’s senior director for Product and AI introduced the potential for agentic AI to transform buildings operation, with new products being introduced to the OpenBlue platform in the coming months. She said the goal was to translate the ‘magic’ provided by AI in every day life into the world of HVAC and building services.

She said: “We are human and humans expect magic…We have seen that type of magic when we get rerouted to avoid some construction work on the road or there is an accident in front of us…We have changed the way we interact with technology in our everyday life… So why wouldn’t we have that same magic when we go to work?”

The OpenBlue platform has been using AI to harness the data from building sensors to provide diagnostics and prediction for a number of years, she said, but the ability of the agentic AI to self-learn will provide new opportunities to optimise performance.

Irina said: “Our airside optimisation solution, which works for air handling units, VAVs, and rooftop units, has been running on an IT campus in Germany and has achieved 8.5% energy savings from only two air handling units. The saving is attributable to AI, because it’s a closed loop optimisation engine that proactively changes setpoints directly in the system on equipment, so that it optimises for energy savings.”

A key benefit from the energy optimising is that the occupants have also benefited from greater comfort, she added:  “You’re not changing the temperature set point and making everybody upset because it’s getting too hot in the room. There have been no comfort complaints related to this work. So that is what I would call success when it comes to delivering AI which is available now.”

Another example is Stanford University, she noted, which has saved half a million dollars per year in energy costs through AI powered optimisation, but this time on the water side, by optimising its chiller fleet.

Benefiting from agentic AI

But the next step is use of agentic AI, commonly seen with the ‘chatbot’, Irina said. This technology, which Johnson Controls calls OpenBlue Intelligence, enables the building user to ask questions of the system, either using selected prompts or simply: ‘What’s my electricity usage from the last month and how is this electricity usage relating to particular equipment that I am monitoring? Or how are the faults related to particular equipment?”

The system has also been successfully used in non-equipment settings, such as occupant capacity, with a number of customers reporting significant gains from better use of office space, for instance.

Irina noted that the technology is not simply ‘dumping data into a chatbot’. She said: “You need to have connections between the different data points established. The system needs to understand that that equipment is related to that space; that is related to that fault; and that it’s related to those alarms that are triggered. And that’s the complexity that we are actually taking care of on your behalf.”

The company describes the concept as ‘using an AI system like you would use a colleague.’ “You can focus on the human intelligence aspects and you don’t need to worry about spending hours and hours writing reports and looking for particular data, or having to understand the correlations between this equipment failing and this particular alarm being triggered.”

A key benefit of agentic AI is to prompt the facilities manager, she noted, giving the example of a chiller system:  “Chillers naturally vibrate. When that vibration changes its pattern over time, you may not notice because most people don’t go and listen to the chiller every day…But that can be measured and even a very traditional models of AI can tell you when that pattern gets out of order and can prompt you to look at the equipment.”

But where the optimisation can really accelerate, she added, is where the AI agent delegates a task to another agent, which is how the chiller and AHU optimisation works: “Essentially you’ve got an AI system that’s now calculating what would happen if it’s making a prediction, based on the data it has today – how that data will impact the system. For example, ‘If occupancy goes to this point, how is the temperature and the air in this room going to be impacted?’

Alternatively, the agent might assign itself a task based on its own decision such as generating the work orders automatically, once a fault has been found- thus saving the need for a technician to diagnose it..

Irina said the way to treat an agentic assistant is to treat it as a new coworker that needs to learn the methods over time:

“We start with a generic model, which is already trained on vast data that we have from around the world, but it will then need to learn to become yours. Sometimes it needs a couple of days. Sometimes it needs a couple of weeks. Sometimes it needs a bit longer – to learn the patterns across the seasons for instance.”

Rolling out OpenBlue Intelligence

Johnson Controls is rolling out OpenBlue Intelligence in a variety of ways – ranging from an add-on capability to monitoring already has in place, right up to the agentic AI OpenBlue building, where the building would be effectively monitoring and even ‘running’ itself, using the platform,

Irina also emphasised that with these new systems, the building owner or operator owns the data and Johnson Controls provides the algorithms.

She added: “You will get the algorithms that make sense…humans remain always in control…and there is always something installed on the premises that keeps the guardrails in place, managing the risk without compromising the benefit.”

The company is introducing the system with modules dedicated to different sectors, starting with mission critical areas such as healthcare, banking and pharma.

She said: “We give you a recommendation for what to do based on the goals and on the data that we collect from the site, and we summarise for you the concrete actions – we prioritise for you what you should be doing. And based on that, you can continue the conversation. You can just say, “OK, I’ve done this already. What should I be doing next?”

An example, she said, was a work cafeteria. If there is a problem with the ventilation, the FM would just have to say ‘there’s a problem here’ and the agentic AI would look for the problem. She said: “Why would you need to pick from a very long dropdown menu and waste all that time? AI can do that and it took less than a minute to report something. That’s how we are transforming this portfolio.”

The new agentic products will be rolled out from Autumn onwards, providing a huge opportunity for buildings to benefit, she concluded: “I encourage everyone not to be afraid, but to embrace the opportunities and remember these agents get better every day. Let’s stop reacting and start acting based on what technology can give us today.”