Utilities have used AI and machine learning for years to improve outage prediction, work planning and asset management. Today, generative and agentic AI are opening new ways to interact with enterprise systems and automate work.
But Barnes cautions that utility organizations cannot simply replace proven systems with AI-generated applications.
For highly regulated, mission-critical environments, that experience remains essential to building solutions that are secure, scalable, reliable and resilient.
Generative AI is making it possible for small teams to prototype new capabilities quickly. CGI's “Skunkworks” approach brings together early-career talent in areas such as AI, programming and design with experienced utility professionals.
The teams have explored use cases ranging from Uber-like customer appointment tracking to AI-powered outage and work-duration prediction.
The goal is to move quickly without losing the engineering discipline required for enterprise utility systems.
Barnes sees particularly strong potential in wearables, mixed reality and multimodal AI. Connected glasses and other devices could give field technicians real-time access to asset information, telemetry, directions and work history without requiring them to move between applications.
AI could also recognize asset conditions in the field and help generate inspection reports or work orders, giving technicians greater awareness while reducing administrative effort.
Another major shift is the emergence of a “headless” experience, where employees interact with enterprise systems conversationally rather than navigating tables, filters and forms.
An AI agent could identify poorly performing feeders, schedule inspections, create work orders and assign them to the appropriate crews—turning natural-language requests into coordinated operational action.
For Barnes, the ultimate value of AI is giving utility professionals better information and stronger situational awareness.
As utilities modernize infrastructure and strengthen resilience, the opportunity is not to replace decades of expertise, but to make that expertise more accessible and actionable through AI.
- 1. How AI in utilities is evolving from machine learning to agentic AI
-
Peter Warren:
Hello everyone and welcome to another installment in our continuing series on how things are changing in the energy industry in all formats. Energy transition, of course, has been a big point of it. AI now is hugely top of mind. Today my guest is Peter Barnes. Mr. Barnes, do you want to introduce yourself?
Peter A. Barnes:
Absolutely, Pete. Good to be here. Longtime listener, first-time caller.
Peter Warren:
Ha ha ha.
Peter A. Barnes:
I'm Peter Barnes. I'm the solution director for CGI's OpenGrid solution. Coincidentally, today is my thirty-fifth anniversary at CGI.
Peter Warren:
Yes, congratulations.
Peter A. Barnes:
Thirty-five years of experience in utilities, whether that's outage management, workforce management, network management, enterprise asset management or distributed energy resource management. So, happy to be here and happy to talk about artificial intelligence and how things are progressing, especially when it comes to the utility industry and its adoption of artificial intelligence.
Peter Warren:
Well, given that this is your thirty-fifth year here and you've got a little bit of a history perspective, this is not our first time with AI. We've looked at AI in a lot of forms over the years, including machine vision. We've also looked at advanced research. I know one of the functions that the solution you have does is event replay; we've looked at using that for advanced training and all sorts of initiatives. Where do you see the evolution starting from where we were to today?
Peter A. Barnes:
Well, we've definitely made leaps and bounds. When we're talking about today, we're talking primarily about generative AI and agentic AI, but where we started from an AI perspective was really things like machine learning, being able to come up with a model to figure out how long durations are going to be for work orders given the past as prologue. So if it took me an hour to do this thing in this part of the city and it took half an hour in this part of the city and it was raining or it was snowing, being able to come up with a model that we would be able to inject the current state into and have that model figure out, it's actually going to take forty-five minutes because it's raining but the winds are southeasterly. So machine learning algorithms have been around for a while.
What's changed and what's evolved in the last, I'd say, pretty much two to three years is the genesis of true generative AI and agentic AI. Generative being able to be the ChatGPT that we rely on for our emails and our translations and everything we do nowadays, recipes and product comparisons and the whole nine yards. And agentic, which brings us an entirely new dimension of how you can interact with software solutions, especially enterprise software solutions like our own.
- 2. Why enterprise AI needs trusted utility data and industry expertise
-
Peter Warren:
We see people talking about replacing existing solutions. I know in regulated industries that's increasingly much more difficult than maybe in a deregulated industry because of all the complexities. I mean, your solution matches health and safety, all sorts of different metrics that would be very difficult, at least in the short term, for an AI solution to replace. So how do you see the connection between maybe that very high-quality, accurate data you have and new use cases for AI on top of it?
Peter A. Barnes:
The new use cases I think really come out of, essentially now that everyone has access to Claude Code and Codex and other generative AI coding tools, the quantity of AI slop and the quantity of vibe-coded solutions is going to be increasing. For those of you who aren't familiar with the term, vibe coding is essentially being able to take a tool like Visual Studio Code, link it into Claude Code or Codex and just say, "Build me an app that does this thing."
From a utility perspective, as an example, you can say, "Build me an app that does damage assessment." And if you're working at the utility, you might think that's pretty cool. It can do the app, it can build a web page, it can accept attachments, it can even show you the details on a map. And that's great. That's vibe coding. You've been in a vibe and you've managed to output something that's usable. It's usable by you. It's not scalable. It's not secure. It's not an enterprise-class application. And that's really where the rubber hits the road for AI coding and vibe coding and the future of solutions such as OpenGrid.
We aren't AI slop; we aren't vibe coded. We're based on a knowledge base and I'm one of the knowledge bases of 35 years of experience, but there are others as well, some coming directly from utilities, some that have grown up in the coding industry and that expertise isn't translatable.
So you can't ask a vibe-coded application to understand the full intricacies of how utilities work for decades. So the role of places like CGI and consulting companies and also IP companies is to bring that expertise and to leverage it using bleeding-edge AI techniques.
- 3. Why resilience and scalability matter in AI-enabled utility systems
-
So it's the meld of that bleeding-edge technology and the generative AI and the coding with the experience in order to not only put out something that's aesthetically pleasing, but to put out something that's technically viable and also is built on an architecture that scales. An architecture that's reliable and resilient, one that is fault tolerant, one that's able to keep large utilities and small and medium and mid-sized utilities running on the worst day of their lives.
You're not going to vibe code an outage management system and then replace the one that has provided you, because all of our expertise, all of our knowledge, all of our testing capabilities have gone into making sure that outage management solution is not only resilient, but it's with you on the worst day of your life.
That's true for field service management and that's true for enterprise asset management. It's all about scalability, but it's about translating the know-how into products that make the best possible use of those bleeding-edge technical solutions.
- 4. How guided AI innovation can accelerate utility use cases
-
Peter Warren:
It's not just our software, but any of these sorts of foundational pieces of code have that same story. So you've also made some very interesting innovation in your group, taking people that know nothing about the industry. You called it Skunkworks after the Lockheed Martin Skunkworks and you created some very cool things because you brought in their expertise. Maybe you could describe that for us as well.
Peter A. Barnes:
Yes, absolutely. Skunkworks is a term, as you mentioned, popularized by Lockheed Martin. One of the outputs of Skunkworks was the SR-71 Blackbird, which is the plane that the X-Men used in their comics a hundred and fifty years ago.
But Skunkworks is essentially just putting together a small group and focusing them on a task. So, taking them outside of the standard flow of things, the approval flow, the getting investment flow, the running the rigorous testing flow and just innovating. So just raw innovation.
So what we did is we took a few early career professionals at CGI, once again who had just come out of the university system, that understood machine learning, that understood graphical design, that understood the use of tools like Figma, that understood artificial intelligence and programming and put a few tasks in front of them.
One task was to build a pizza tracker application. Essentially, as a customer of a utility, if I've scheduled work, I'd like to have an Uber-like experience where I know when that person is going to show up, I can track them on a map, I can see and communicate with them bidirectionally and say, "I'm actually not going to be around." I can fill out pre-visit checklists and if they aren't filled out, then from a systems perspective, I can turn around or reschedule or move to my next assignment.
So this customer-to-utility interaction, and this is of course something that translates to telco, it translates to, in my case, Videotron, which is the local provider of internet. I don't want an AM/PM appointment. I'd like to know when you're on your way so I can plan my days accordingly. So providing that Uber-like experience to customers was the first output of the Skunkworks team.
- 5. How AI can improve outage prediction, forecasting and field efficiency
-
And the second output was, again taking into account all of our historical information, we created some synthetic data that mirrored what our actual customers go through. And using that synthetic data, came up with a model to allow us to predict outage durations and work durations and estimated time of restoration. And again, these are people who did not have the grounding in the utility industry.
But because they were provided with the information from CGI's experts, from myself and from others, they were able to rapidly deploy prototypes that then can go into a rigorous product cycle. They can become part of the product roadmap. So, it's innovation and it's guided innovation, but it's also taking into account all of that experience, not just randomly popping up bits of code without any kind of governance framework.
It's something that we found very useful and something that we're looking to continue. The Skunkworks team will continue with some new initiatives, for the most part all having to do with the application of machine learning and artificial intelligence into the OpenGrid applications, whether that's from a prediction perspective, a forecasting perspective or just being able to get more efficiency out of the field technicians by optimizing their routes additionally through the use of artificial intelligence.
- 6. How multimodal AI and wearables could transform utility field work
-
Peter Warren:
Where do you see voice, wearables, multimodal AI coming into this?
Peter A. Barnes:
The multimodal aspects, the mixed reality aspects, those are the ones that actually excite me the most. They're essentially three degrees of freedom lenses or glasses that allow you to broadcast any content from an iPad or an iPhone or a MacBook into a mixed reality space. So it's essentially like having enormous monitors taped to your eyeballs. It's wonderful as an experience, especially for consuming media.
But the next generation, I think it's called Project Aura, the next generation of this is actually going to use mixed reality using Google Android XR to allow you as a human being to interact with, let's say, six degrees of freedom. So you can move around objects that you place in space and you can look at something and let's say interrogate it.
And if we look in our world, interrogating something might be looking at a transformer or a switchgear and having real-time telemetry from SCADA, real-time telemetry from the ADMS or OMS, to have that updated to see what the state of actual devices are, or to interrogate about a premise where you're going to actually do some work, pull up all of the metadata that we know about that house, all of the history and have it in essentially the heads-up displays that we've been promised since Terminator 2.
Being able to have those be real and use those not just as a curiosity, but as part of the day-to-day operation of field technicians. So for me, this is a very exciting time in wearables because a renewed focus on the Meta glasses, the XREAL glasses, to provide a user experience that's seamless, but also one that's leveraging real-time information.
So give me real-time directions, show me an arrow to point me in real time towards my next assignment. Or look off into the distance and see where my next assignment is, generally 1.2 miles away from here and have something hovering in space that shows me approximately where it is.
So all that to say that wearables, when we're talking rings, we're talking watches, we're talking the glasses, whether they're performing monitoring of a person's health, which is amazing, whether it's performing real-time monitoring of assets and work and crews and things in the field, or whether or not it's giving you notifications about the world around you as you're driving by, it's actually looking at the poles or it's looking at the transformers as you drive by and identifying conditions almost subconsciously, so that you don't even have to be looking at it.
Also, obviously the privacy implications are absolutely insane, but hopefully the necessary regulations, the necessary framework and guardrails will be in place to prevent abuse of these technologies. But one thing's for certain, they are coming—ready or not, so it's best to be ready.
- 7. How agentic AI can improve situational awareness, efficiency and field safety
-
Peter Warren:
For over a decade now we've been doing something similar in the aviation industry, using virtual reality and things and actually allowing people then to have the best engineer, maybe even the manufacturer themselves, over their shoulder saying, "Well, that engine isn't working because of this." So helping the technician in the field, where do you see the very next future where technicians will be able to do more work, safer work, more diverse work because of this technology?
Peter A. Barnes:
It all boils down to essentially situational awareness for me. Knowing what's going on and being prompted to know what's going on before finding out what's going on.
So as I'm heading to a new customer, that there is an agent that's validating safety. As I'm going into a new neighborhood that I've never been before, that there's an agent that says, "Well, this is one thing, two blocks over the last time somebody got chased by a dog."
Again, using proximity, using spatial awareness of where you are and where you're going to be in order to drive information being streamed to the field technicians and also obviously to the back office as well.
- 8. How agentic AI could transform utility workforce management
-
The headless capabilities, so being able to have a user experience that's more voice-based, that's less of a tabular list of things that I have to look at and sort and filter and more of a conversational, "Good morning, Samantha." (We call one of our AIs Samantha).
"Good morning, Samantha, what's on the docket for today?" And it understands that I use the word docket because I was a Law & Order fan and it's figured out that it just means, tell me what the plan is.
And it's learned my areas of interest. So, I'm interested in maintenance work and of course in emergency outage restoration work. It'll then itemize those and it'll prioritize those, give me a list and give me my to-dos, the things that require attention and also summarize the things that the agents have taken care of while I haven't been around.
So the headless experience for me is where I see the most potential disruption in the way that work is done and the way that work is tracked, because we are used to pull out that tabular list and create your column filters and filter it by the agency or show me only these date ranges, right? And we're used to configuring something like that. Whereas an interaction now can simply be speaking it and having the AI, having the agent take the appropriate steps.
"Create for me a new job" or "check the feeders that are my worst-performing feeders and schedule inspections for the next six months on those feeders, but only use the top ten of them." And to have the AI, again the agents, then go out, figure out what return they should get, but also start the process of creating the work orders, assigning the work orders to the skill sets of the appropriate crews, scheduling all of that work in the future for whatever time horizon it takes.
All of that can now happen in a state that doesn't require extensive use of tables and detailed forms, et cetera. We're really at the cusp of that headless experience.
- 9. The future of AI in utilities is human augmentation
-
Peter Warren:
Combining the two, you mentioned earlier about the situational awareness where it sees something. You could be in a conversation where it notices that something needs maintenance, it needs fixing. Something's not right. There's a wire out of place or something. It could then bring that to your attention, have a conversation with you whether you're equipped to fix it now or not and create the work order for tomorrow or another day.
Peter A. Barnes:
Exactly.
Peter Warren:
Really these are not something to fear in the field. This is really a multitasking, multi-reality tool for you to be able to share information and hopefully do better and safer work for years to come. It's not replacing people, it's really bringing their expertise to the forefront, would you say?
Peter A. Barnes:
Yes, absolutely. It's augmenting their situational awareness, augmenting what they can do. Obviously, it increases efficiency, but it's providing you with essentially superhuman awareness of your surroundings, which is pretty awesome.
Peter Warren:
That's a great point to end on, Peter. Thank you very much for your time and expertise. And thank you, everyone.
