AI and Power Engineering — Part 1
Like many engineers, I have been interested in both the potential and the limitations of AI for some time. In the early days it was often superficial and unreliable, getting basic concepts wrong, muddling things like reactive power, and hallucinating far too freely to be trusted on anything technical. That landscape has now shifted quite dramatically. Modern agentic AI systems are becoming far more capable, particularly when grounded in the right technical context and tools.
Over the next few weeks, I will be sharing a short series of posts on where AI can genuinely add value in power systems analysis. Not as a gimmick, and not as a simple question-and-answer tool, but as a way of adding more depth, speed and technical insight to real engineering workflows.
The popular framing is usually something like this: can AI write reports, summarise standards, or answer technical questions on subjects we are not familiar with? That is probably the most common use case, and whilst helpful, it is not where the real value sits.
Where the value really is
In power systems analysis, the hard part is rarely just producing words. The hard part is navigating the model, understanding what matters, interrogating assumptions, connecting the right pieces of technical context, and applying judgement in a way that is fast enough to be useful without losing rigour.
That is where modern AI starts to become genuinely interesting. When paired with structured technical information, model-aware tooling, and a well-designed interface layer, it can help an engineer move through a complex study space more effectively. It can surface relevant context, organise fragmented information, and help build a more coherent picture of what the model is doing and why.
Not a replacement for judgement
This is not really about replacing engineers, nor is it about handing critical decisions to a language model. It is about using AI as an interface to technical knowledge and model structure in a way that supports engineering judgement rather than pretending to automate it away.
That distinction matters. Generic AI output, without grounding, is still often weak. But AI combined with the right tools, structured inputs and direct model interrogation becomes much more useful. The difference between those two modes is enormous, and I suspect that many of the most valuable applications in power systems will sit squarely in that gap.
What comes next
In later posts, I intend to look more closely at where this can matter in practice: reduced network models, study workflows, model interrogation, structured knowledge layers, and the ways AI can help make technical analysis both faster and more insightful without becoming careless.
That is the area that interests me most — not AI as a novelty, but AI as a serious engineering interface.