“Hey, why do I need Excel?”
Microsoft’s Satya Nadella posed that question earlier this year, pointing to a future where AI agents will replace traditional software.
In tech, it’s a rallying cry. But in utilities, it prompts a raised eyebrow. In an industry where mistakes can mean outages, blown transformers, or worse, the idea of outsourcing complex decisions to autonomous agents deserves a closer look.
Utilities today are navigating a landscape defined by uncertainty. As one industry executive recently told me, “We’re in the most challenging planning environment in decades.”
We all see the growth curves — electrification, onshoring, and data centers — but exactly where, when, and how much load growth we can expect remains highly uncertain. And if history has taught us anything, the risk of overbuilding is very real.
Now, layer on top of that growing rate complexity, distributed resource integration, reliability mandates, shifting customer expectations, and lengthy planning cycles. Add inflation, regulatory pressure, tariffs, and capital constraints. Every part of the enterprise, from generation to T&D to customer operations, is feeling the squeeze.
And as a result, rates are heading up. In New York alone, regulators processed more than 60 full-year rate cases in 2023, up nearly 40% from 2021 levels. And that’s before the wave of 2024 and 2025 rate increases are taken into account.
For all these reasons, utilities are in transformation. The question is: how can utilities navigate the promise and pitfalls of AI at this inflection point?
The necessity of skepticism
If you believe the pitch, AI agents are the answer to every problem: software that can reason, decide, predict, and act across your enterprise. Planning? Load forecasting? Customer outreach? Asset management? Let the AI handle it.
A utility CIO told me he gets these pitches 10 times a day. But he and his many peers are right to be skeptical: both because it’s challenging to apply probabilistic thinking in a deterministic world, and the complications of heavy, expensive solutions.
Utilities are built on engineering rigor, on systems that need to work at all times.
When I first entered this industry, I thought the culture was overly cautious. Then I realized: Utilities are literally in the business of not blowing up houses, of sending lineworkers into dangerous conditions to keep the grid up, of staying out of the news, because that means everything’s working.
AI, at its core, is probabilistic. It makes predictions. It’s not always right, and it’s not always clear when it’s wrong. Integrating AI into the utility environment means managing a whole new layer of uncertainty. Utility-grade AI requires industry expertise and clear guardrails: human oversight, audit trails, and an ‘I don’t know’ button built in.
This may take time. Today, AI language models sometimes sound confident while being dead wrong. The industry is working hard to improve hallucination rates in LLMs, but the tooling and workflows for identifying and mitigating those errors are still nascent.
Another challenge is that many AI offerings today are old wine in new bottles: legacy SaaS platforms wrapped in “AI” branding, protected by consultant-heavy business models.
This means multi-year timelines, multi-million-dollar invoices, and fixed architectures that can’t adapt to changing realities. We’ve seen projects that should take months today stretch into years — not because of tech complexity, but because of delivery inertia.
That said, there’s reason to be excited anyway.
AI is the Gutenberg moment for computers
LLMs are doing for computers what the printing press did for people. Software can now read, hear, talk, and see at significantly lower costs than just a few years ago.
Utilities sit on mountains of unstructured data, including spreadsheets, PDFs, microfiche (yes, still), and handwritten notes. Unlocking that data used to take armies of consultants, or it would just sit unused, orphaned off in a PDF black hole. Now, a well-tuned AI agent can handle it quickly and turn it into value.
Case in point: PG&E recently deployed a generative-AI “records copilot” at its Diablo Canyon nuclear plant, the first on-site gen-AI deployment at a U.S. nuclear facility, to help engineers instantly search millions of pages of decades-old licensing and operating documents.
You don’t have to believe in artificial general intelligence for this. Instead, the technology can just be a better way to read what we already have.
At my previous company, EnergySavvy, we delivered a Next Best Action customer optimization platform to personalize utility customer experiences, driving up customer satisfaction and optimizing costs. This used analytical AI techniques and data wrangling that are tried and true.
But, with AI, we can do much better — and do it faster and cheaper.
These systems can interpret complex data and adapt to various roles and contexts, such as a customer, customer service representative, engineer, field technician, or trade ally.
They can also deliver solutions that are more cost-effective, work across data silos, all while being easier to capitalize and configure around business processes, unlike rigid, traditional SaaS tools. Oh, and we can do it in many languages to help more people more easily. Qué bien!
But we can’t rely on AI alone. One of the most powerful ideas in modern AI is augmented intelligence: human and AI collaboration. Studies show human-AI teams often outperform either humans or machines working alone.
We already see this in California’s ALERTCalifornia wildfire camera network: An AI agent flags possible smoke from more than a thousand remote cameras, but human dispatchers validate the alerts before crews roll. The result is faster triage with human judgment in the loop.
For utilities, this means using AI to reduce manual toil, generate insights across disparate data silos, and suggest paths while humans validate, correct, and lead. This isn’t about replacing planners or operators. It’s about scaling their intelligence and freeing them to focus on higher-value work.
Making the technology work for you
We’re early in the journey of adopting AI for utilities, but some steps are already clear:
1. Build AI literacy across the org
Your teams are already the smartest people on the problems they face. Help them become fluent in what AI can do, where it fits, and where it doesn’t.
At AZX, we’ve seen incredible unlocks happen quickly when sitting down with cross-functional utility leadership teams. Once domain experts understand the full spectrum of AI tools, from LLM-based agents, to digital twins, predictive analytics, and computer vision, their creativity accelerates and new pathways to value emerge.
2. Pick a direction: forward or backward
AI won’t stay optional for long. What feels like innovation today will be table stakes tomorrow. Regulators, customers, and market forces will expect AI-forward capabilities. The risk isn’t moving too fast; it’s getting left behind.
3. Start small, but start now
Start the next pilot. Draft the business case. Build in human oversight. Iterate. Some of the most valuable wins come not from moonshots but from unglamorous problems solved with smart AI initiatives that work towards a strategic north star.
Entergy, for instance, reports that an AI-and-satellite vegetation management program helped beat reliability targets by more than 30% while holding budgets flat. This is the kind of bounded, measurable use case where utilities can start small and see real results.
At the end of the day, it isn’t as simple as “adopt AI agents or don’t,” or even “use AI here and not there.” These choices will be unique to each utility, the operating environment, and the challenges ahead.
Hype? Yes, plenty. Promise? Absolutely — when applied thoughtfully. We all have the opportunity to get smarter, deliver more quickly, start small, and focus on results, not the noise.
Aaron Goldfeder is the CEO of AZX. He previously was the founder and CEO of EnergySavvy, which was acquired by Uplight. The opinions represented in this contributed article are solely those of the author, and do not reflect the views of Latitude Media or any of its staff.
