Cart
There was a time when using software meant telling it exactly what to do: click this button, enter this value, run this report.
Artificial intelligence is beginning to change that relationship. Instead of navigating through pages of information to find what we need, we can increasingly ask for it. Instead of searching a hundred-page technical document for exactly the right term, we can describe what we’re looking for. And instead of writing every line of code ourselves, we can ask an AI tool to help build it.
For building automation professionals, that opens some interesting possibilities. It also raises an important question: If AI can do more of the work, what happens to the expertise behind it?
The answer may be somewhat counterintuitive – human expertise could become more important than ever.
Getting to the Information Is Getting Easier
Building automation systems already generate enormous amounts of useful information. The challenge has often been getting to it.
A technician may need to navigate multiple systems, search documentation, examine trends, compare points, or dig through data before reaching the information needed to solve a problem. AI has the potential to shorten some of that process considerably.
During a recent Stuck on a Bucket webinar, KMC Controls Software Development Manager Cody Geisler demonstrated one example using KMC Commander®. With read-only access and KMC Commander’s open API documentation, an AI tool was able to retrieve trend data from an energy meter and interpret it without Cody having to specifically program the interaction himself.
That’s a glimpse of a much larger shift. The value isn’t necessarily that AI suddenly knows more about the building than the people operating it. It’s that getting from a question to the relevant information may become much faster.
Start With the Boring Stuff
Ask someone where to begin with AI, and it’s tempting to look for something impressive. Cody’s recommendation during the webinar was much more practical: start by using it to summarize complex information.
Have a hundred-page technical document and can’t remember where a particular input is described? Ask AI to help you find it. Need to understand a specification but aren’t sure you’re searching for the right terminology? AI can often help find information based on related concepts rather than relying entirely on an exact keyword match.
Mechanical prints and other visual documents are becoming more accessible to multimodal AI tools as well. Cody suggested that if uploading an entire document doesn’t produce a useful result, taking a screenshot of the relevant portion and giving the AI the image can sometimes work better.
None of that sounds revolutionary, and that’s partly the point. Before asking AI to control anything, there is a lot of value to be gained simply by using it to find, organize, summarize, and explain information faster. But as AI moves from helping us find information to interacting more directly with software and systems, security becomes part of the conversation too.
Doing the Work Isn’t the Same as Understanding the Work
Things get more complicated when AI moves from finding information to creating something. Software developers are already experiencing that shift. AI can generate functions, write tests, and produce significant amounts of code in a fraction of the time it would take a person to type it manually.
But somebody still has to know whether that code is any good.
Cody described an interesting change in his own work: developers can begin to move from being primarily code writers toward becoming architects and reviewers. Instead of spending all their time creating the individual pieces, they spend more time directing what should be built and evaluating what the AI produces.
There’s a lesson there that extends well beyond software development. AI can produce an answer remarkably quickly, but reviewing that answer responsibly may require considerably more knowledge.
The Technician Still Has to Know the Building
Imagine an AI tool looking at a building’s trends and identifying something unusual. That’s useful, but is the unusual condition actually a problem? Is there an operational reason for it? Does the recommended change make sense for this particular piece of equipment, sequence, facility, or application?
Those questions require context. The person working with the system understands things that may not be represented in the data. They know the equipment, the application, the customer, and the consequences of making a change.
AI can assist with that work, but it doesn’t automatically inherit that expertise. As Cody explained during the webinar, the important question is whether you’re simply letting AI do what it thinks is best or whether you’re the system expert driving the architecture.
That’s an important distinction.
Capability Isn’t the Same as Trust
When asked what he wouldn’t trust AI to do, Cody’s first answer was simple: life-safety or other critical work.
That doesn’t mean AI has no place in those environments. It means the level of review needs to match the consequences of getting something wrong. The same principle applies elsewhere: the less qualified you are to evaluate an AI-generated answer, the more cautious you should be about acting on it.
AI can make it surprisingly easy to produce something that looks authoritative. That can create a dangerous illusion: if the AI knows how to do something, perhaps we assume we know how to do it too.
We don’t – and when software begins interacting directly with physical building systems, that distinction becomes particularly important.
More Capability Means More Attention to Security
The same capabilities that make AI useful can also introduce new risks. As Cody pointed out during the webinar, AI is making sophisticated tools more accessible – not only to the people using them for legitimate work, but potentially to those looking for vulnerabilities as well.
That matters as AI becomes increasingly capable of interacting with software and connected systems. Cody demonstrated how an AI tool could use KMC Commander’s open API to retrieve building data without him having to specifically program that interaction. That accessibility can be incredibly useful, but it also reinforces the importance of controlling who and what has access to a building system, what they are allowed to do, and how that activity is secured.
AI itself can introduce another layer of risk. Cody showed how seemingly harmless information presented to an AI could potentially steer it toward an unsafe action, such as directing a user to a fraudulent email address. As these tools become more capable, network security, user permissions, secure access, and careful review become increasingly important – not less.
For building automation professionals, cybersecurity is already part of protecting connected buildings. AI gives us another reason to make sure those foundations are strong.
A Different Kind of Productivity
It’s easy to imagine the future of AI in building automation as technicians simply doing the same jobs faster. That may be part of it, but something more interesting could happen.
Tasks that previously weren’t practical because they required too much time may become practical. Technicians may be able to explore more advanced control strategies, investigate problems more deeply, build better testing processes, or spend less time hunting for information and more time applying what they know.
Cody pointed to research and communication as areas where there are already significant efficiencies to be gained, even before some of the more advanced automation possibilities become commonplace. That doesn’t eliminate the role of the expert. It changes where the expert spends their time.
Keep the Human in the Loop
AI in building automation doesn’t have to be a choice between embracing every new capability and refusing to trust any of it. There’s a much more useful place in between.
Use AI to find the information, organize the data, suggest possibilities, and handle some of the tedious work. But give it appropriate access, protect the systems and data it touches, and review what it produces. Then bring the thing AI doesn’t have to the process: your experience.
Because as these tools become more capable, knowing how to ask them to do something will certainly matter. Knowing whether they did it right may matter even more.
Watch the whole episode here: