AI’s Role In Energy Isn't Replacing Experts—It’s Extending Their Reach
Guilherme Studart is the co-founder of Delfos Energy, an AI Asset Performance Management solution for Wind, Solar and Storage.
gettyFor years, the most valuable diagnostic instrument on any wind or solar site wasn’t software. It was a seasoned technician who could read a small drift in the data and know it meant a gearbox and not a faulty sensor. That judgment took decades to build, and it was mostly unwritten. Now, across our industry, it’s walking out the door.
Our performance engineers have spent years working with wind turbines, and just as long figuring out what’s really behind a dashboard or a chart. It takes hours of investigation: forming hypotheses, debating, ruling things out and eventually finding answers. I’ve spent the last nine years doing this alongside our engineers, and we logged everything. Most asset managers don’t have the resources to do that. And the ones who do rarely write it down. That knowledge lives in someone’s head, and nowhere else.
The International Energy Agency (IEA) found that “in advanced economies, there are 2.4 energy workers nearing retirement for every worker under 25” entering the field, and that between 2025 and 2035, “two of every three new hires will be needed just to replace retiring energy workers.” In IEA’s 2025 survey of “700 energy-related companies, unions and training institutions, more than half reported critical hiring bottlenecks.” We’re trying to operate rapidly growing renewable portfolios with an aging workforce while demand for skilled workers grows.
The instinctive reaction is to assume artificial intelligence (AI) will replace those technicians. Instead, you need to determine how much of the departing expert’s judgment your systems can carry, so the humans you still have are freed for the work only humans can do. The model I see emerging is neither AI instead of people nor people without AI. It’s a partnership, and the useful way to think about it is as four levels of trust.
The system notices that something is anomalous and raises a flag. Much of the industry still operates at this level, and it’s useful because no one can watch thousands of signals across a fleet without tiring. But detection alone does almost nothing for the labor problem because an alarm still needs the expert to interpret it. You haven’t reduced your dependence on that person. You’ve moved it to a screen and added an alert.
The system not only says something is wrong, but also what and why. This is where the partnership starts to pay because diagnosis is what the retiring engineer used to own. The machine proposes the cause across more data than a person could hold in their head, and the expert confirms it and decides what to do. Their scarce time now goes to the decision and not to the detective work.
The system proposes the action, the timing and the tradeoff, and the human approves it. Curtailment is a concrete example. As renewable penetration rises, more generation can be curtailed because the grid can’t always absorb or transport it when it’s available. In Brazil, “solar curtailment reached 14% in 2024,” according to reporting in pv magazine. When you can anticipate those windows, the system can recommend scheduling maintenance into them, so that some otherwise forced idle time can become planned maintenance time.
This is the level at which the system executes within boundaries a human has set. Here’s what that looks like in practice. At one of the wind farms we work with, 28 anomalies were flagged that sat below SCADA alarm thresholds: rising temperatures in generators, gearboxes and transformers, plus miscalibrated nacelle anemometers. Nothing was dispatched automatically. Our engineers investigated each signal and built an action plan with their maintenance team, who decided what to fix and when. Corrections were completed in about 45 days, avoiding a projected 9,629 MWh in losses over one year.
However, relatively few operational environments are ready to delegate consequential actions without tightly defined human controls. And I’m wary of anyone who claims otherwise.
And you can’t skip levels. So, for each system you rely on, be honest about which of these four levels it truly reaches. And consider whether it frees your scarce experts for higher judgment or simply relocates your dependence to a dashboard.
In short, the goal isn’t to replace people but to secure the knowledge of seasoned technicians and asset managers and build on it. That has to happen now, while experienced engineers are still onsite, by putting systems in place that capture what they know before it’s gone.
The four levels aren’t a finish line. They act more as a necessary foundation, because trust in a system is built the same way trust in a new colleague is: one proven decision at a time, with the right context behind the data.
And the stakes go beyond keeping wind farms running at peak performance. Every turbine that stays online is part of how we secure our own energy.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
