AI can write, plan, and predict but can it run the grid? This piece explores where it’s already adding value, where it still fails, and why the future of a “self-healing” grid will depend as much on people and infrastructure as on algorithms.
I’ll admit it. I’m a little obsessed with AI. “Let me check ChatGPT” has basically become my mantra. I use it for everything: writing, playlists, workouts, recipes, even taking apart a trampoline once (which it somehow nailed). It gives advice, encouragement, and the occasional ego boost. What more could you want from a machine?
There’s no doubt it’s powerful. But can a technology that can’t even produce the same image twice, or remember to stop using long dashes (—) when you ask it to, really be trusted to help run something as critical as the power grid?
That’s the part I’m not so sure about. So let’s take a look at where AI genuinely adds value and the challenges it still has to overcome to earn its place in managing a national, security-critical grid.
Artificial intelligence is already reshaping our daily lives and work:
It predicts what we want to watch Netflix recommendation algorithms drive over 80 percent of viewing choices.
It helps us write, plan, and summarise with AI tools like Microsoft Copilot now used by nearly 70 percent of the Fortune 500, and that’s before you even count the reach of Google Gemini or ChatGPT.
Across industries, AI has moved from experiment to everyday utility, boosting productivity, automating tasks, and informing decisions at scale. But can it deliver the same value in the mission-critical world of energy, a system built on near 100% reliability, security, and decades of institutional trust?
Unlike social media, e-commerce, or customer service, the grid isn’t a sandbox or a low-risk playground on the end of a bell curve; it’s a national backbone. So before we celebrate algorithms running our power networks, we have to ask: can AI thrive in a world where failure isn’t just inconvenient, it’s unacceptable? And beyond performance, what purpose should AI serve in the energy system to optimise, to predict, or to guide us toward a more resilient, conscious kind of intelligence?
Most utilities don’t suffer from a lack of data; they suffer from the wrong kind of data. Smart meters, DERMS platforms, and IoT sensors now stream billions of readings, but too few are structured, time-aligned, or labelled for model training. Forecasting models depend on clean, contextualised data; yet the reality is patchy telemetry, missing metadata, and privacy firewalls that stop signals at the substation.
The result: models that look intelligent in test runs but stumble when faced with real-world noise, latency, or hardware faults. In other words, the grid doesn’t need more AI — it needs better plumbing.
According to Accenture, 74 % of utility executives believe AI’s full potential can only be realised when it is built on a foundation of trust, which can only be built on solid data foundations. We seemingly have more sensors than ever, but half the time we don’t trust what they’re telling us.
As renowned MIT computer scientist Michael Stonebraker once said, “Without clean data, or clean enough data, your data science is worthless.”
Interoperability was a central theme in Metering 2.0: Smarter Promise, Familiar Problems. The data flowing from today’s devices is only as valuable as the systems that standardise, clean, and contextualise it. Until those foundations improve, even the smartest algorithms will be building on sand.
AI’s promise in grid management is often framed as prediction and optimisation “predict load,” “optimise dispatch,” “forecast solar variability.” But the true measure isn’t whether the model runs; it’s whether it earns its keep. For most utilities, the ROI of AI depends on the last 10 percent of performance: reducing forecasting error from six percent to five might look impressive on paper but may not justify the data-integration cost, compute spend, or compliance overhead.
We can make a model 10, 20, even 30 percent more accurate but if it costs ten times more to maintain, what’s the real value? It’s a question echoed by Google researchers in their Hidden Technical Debt in Machine Learning Systems
study, which found that marginal accuracy gains often create “massive ongoing maintenance costs” in real-world ML environments. Accuracy may impress on paper, but resilience and maintainability are what actually deliver lasting intelligence.
Before we talk about advanced intelligence, we need basic awareness. You can’t optimise what you can’t see, and you can’t coordinate what you can’t control.
The biggest near-term impact doesn’t come from neural networks; it comes from visibility and control. A universal device registry one that tells us what’s connected, where it sits, and how it behaves would unlock more real flexibility than any black-box model. The ability to reliably signal, aggregate, and coordinate distributed devices is the real superpower that most utilities still lack.
It’s the same story I explored in Cablegeddon: The Revolution We Didn’t Know We Needed. Without standards and interoperability, every new “smart” device just adds more complexity. The grid’s intelligence won’t come from a single AI model but from connected systems that can talk to each other clearly and securely.
The foundation of an intelligent grid is reliable awareness of assets and data. The DOE’s Grid Modernisation Initiative highlights tools to measure, analyze, and predict grid behavior, anchored in interoperable data and device visibility, while NREL research emphasize machine-readable information models and metadata so devices can be identified and understood by machines.
Once that foundation is in place, the rest becomes far easier. Forecasting improves because the inputs are accurate. Automation scales because the endpoints are known and trusted. And AI finally has something meaningful to work with clean signals, consistent context, and a controllable system.
We don’t need to be really smart just yet. We just need to be aware, connected, and ready for intelligence to matter.
AI delivers when the problem is well-bounded, repetitive, and data-rich. Across DER and grid operations, here are ten who look like they are adding real value:
I know from experience that deploying technology in the wild is full of nuance, testing, and refinement. We should take press releases with a pinch of salt and keep an attitude of optimistic caution. AI can misfire in energy, but the biggest failures usually come from system design, data quality, and incentives rather than the model itself.
In my view, the next real step for AI in energy won’t come from dashboards or chatbots it will happen in the darker, harder-to-see parts of the grid. These are the blind spots where data is scarce, latency matters, and milliseconds can make the difference between a local fault and a cascading outage.
Here’s what I think that future could look like:
Whether those actions ever become fully automated is still up for debate. What matters most is that they operate in a trusted, explainable, and secure way.
That’s where I see the future heading: intelligence distributed to the edge fast enough to protect the system, but still anchored in human oversight and security by design.
Even as AI systems grow more capable, human oversight remains non-negotiable for many utilities currently. What we’re seeing emerge is a multi-agent dashboard approach humans in command, AI agents assisting. Each model becomes one voice at the table, analysing data, forecasting scenarios, or highlighting anomalies, while operators still make the final calls.
This hybrid model feels less like a replacement and more like a collaboration. Machines handle the scale, speed, and pattern-matching; humans bring intuition, responsibility, and moral weighting. As one system operator told me recently, “AI can tell me where the storm is coming from, but I still decide when to close the switch.”
That might not sound like the self-healing grid we were promised, but it is a much safer and more realistic step toward it.
By day, Matt Wapples helps utilities and customers unlock the power of distributed energy resources. By night, he blends curiosity, logic, and a touch of chaos to make sense of technology, markets, and the future of power.
His perspective is built on deep experience across the energy and technology sector. He has worked for a utility, advised through consultancy, and helped scale distributed energy platforms internationally. Alongside this, he completed an Executive MBA at Imperial College London, adding a strategic and global lens to hands-on industry knowledge.
Beyond energy, Matt has always kept one foot in exploration and entrepreneurship. He taught scuba diving in Thailand during a year out, and ran Feiyue sneakers into Europe, securing partnerships with major retailers including Urban Outfitters.
Life and work have taken him across London, Germany, Thailand, Singapore, Tokyo, and now Atlanta. That journey has shaped a global perspective and a restless curiosity. Conscious Intelligence (CI) is his way of connecting those dots exploring how culture, technology, finance, energy, and intelligence collide to shape the systems we all live in.
Follow his work on LinkedIn
Nikola Tesla never let go of the question: if music rides the air, why not power? The story traces his early experiments at Colorado Springs and Wardenclyffe and shows how that vision echoes today in wireless charging, inductive transfer, and cautious power-beam trials
A drawer once full of tangled cables gave way to a single universal connector. The piece looks at how USB-C reduced friction, sped up charging, and what its success teaches about the value of standards, from consumer tech to home energy.
Smarter meters on paper, stubborn results in practice. We explain what AMI is, what ‘2.0’ promises, and why those expectations clash with reality. We unpack where advanced meters trip in real homes and why a partner-first approach works better than cramming everything into one box
