Recently, I found myself in a position where I had to present an intelligent perspective on “using artificial intelligence (AI) to improve the (generation) interconnection (GI) queue” from the vantage point of an electric transmission utility. Admittedly, this topic initially gave me pause; it just seemed like another instance of hype-fueled innovation theater that distracts us from addressing the real, underlying problems. However, as with any assignment, I approached it with the professional diligence I would apply to any request from a respected superior or customer. Ultimately, my investigation led me to identify four areas of impact of AI for GI. I’ll begin with some necessary cultural and philosophical framing and finish with a description of each of these four areas of impact. Skip ahead to the impact analysis if that’s your style.
Is This a Valid Question?
Before diving into the question of applying AI to the generation interconnection queue, I first need to validate that the question itself is worth asking. I can’t pinpoint exactly when or why I became sensitive to this, but I find it frustrating—and largely unhelpful—when vendors or others pitch a solution as “AI” or “AI-enabled” as though that label alone will make it more compelling or effective. In reality, I’m not interested in whether a solution uses AI; I’m interested in understanding what specific problem it’s designed to address. The problem should always come before the solution.
With this mindset, any discussion on AI’s role in the generation interconnection queue must begin with a problem-focused conversation. It’s not enough to simply say, “the queue is inefficient” or “it takes too long.” We need to dig deeper, critically examining the queue’s structure, comparing its performance against established expectations, and using any identified gaps as a basis to determine if and where investment is warranted.
Curiosity: Engineers and New Materials
As I reflected on this challenge, two core values from engineering culture came to mind: curiosity and simplicity. I’ll start with curiosity. I’m reminded of a quote from a utility executive in the mid-to-late 2010s, which I find particularly relevant here (while I unfortunately can’t remember his name, the informal nature of a blog fortunately frees me from the obligation to spend too much time finding it). He said something to the effect of, “The most exciting time for any engineer is when a new material is discovered, as it allows them to reimagine every previously created object and explore what new objects can be made from this material.”
AI, in many ways, serves as just such a “material.” Metaphorically speaking, it’s a new, digital fabric we can use to stitch the world together in different and interesting ways. Therefore, we are justified in asking the question, “Can we improve the generation interconnection queue using AI?” much like someone was once justified in asking, “Would this bread and jam taste better with a bit of peanut butter?” Working within a strong engineering culture, I, too, am curious of the potential of new materials.
Simplicity: K.I.S.S.
However, the creation of new materials or reimagining of existing structures does not—and should not—always result in change. Nor should it necessarily lead to the dismissal of “old materials” or the established ways of doing things. This principle, too, is grounded in engineering culture. I vividly recall my first day of engineering school, during my freshman year at Virginia Tech in the fall of 2005. Seated in Burruss Hall Auditorium with ~1,100 other first-year engineers, we listened to an experienced engineer alumnus share his hard earned wisdom. Of all the advice he gave, one lesson stood out: never forget the K.I.S.S. principle, which he translated as, “Keep It Simple, Stupid!”
This principle exists in engineering culture for two primary reasons. First, the simplest solutions are often the most efficient, robust, stable, and cost-effective, making simplicity of design a core virtue. Second, engineers, by nature, tend to over-engineer solutions. It’s almost instinctual. The discipline of simplicity is what helps us transcend that tendency.
Values & Purpose
While these values stand strong on their own, they become even more critical when something important and meaningful is at stake. For instance, in the energy sector, electric utilities face unprecedented, transformative, external forces that demand equally profound internal adaptation. Data center load growth, the clean energy transition, electrification of transport, competition for talent with tech companies and renewable developers, a tightening regulatory landscape, and increasing NIMBY opposition—all in addition to the core responsibilities of safety, reliability, and affordability—continuous transformation is not a nice-to-have but rather table stakes for survival in the coming decades.
You might say, “Well, Kevin, that’s exactly what we’re advocating for! Artificial Intelligence IS transformative and innovative! You’re making our case for us!”
To which I would respond, “In a world where adaptation is essential for survival, the careful selection of adaptation mechanisms is just as crucial, if not more so, than the execution of the adaption itself.”
Put simply, I can’t invest in a transformative technology without a high degree of confidence in its long-term, guaranteed applicability. Think of it like an evolutionary trade-off: every adaptation costs the organism something, and if that adaptation doesn’t directly enhance survival, competitors that didn’t invest in the same adaptation will have the advantage when it comes to “lodging their genes farther into the future.” Similarly, if I invest too heavily in an over-hyped, “transformative” technology at the expense of essential developments, I risk missing the chance to evolve the organization in ways that truly matter. In such a scenario, doing nothing at all may be the better choice.
Aggressive Incrementalism
This is not an argument for techno-conservatism, nor is it inherently a conservative viewpoint. Rather, it reaffirms the fundamentals behind classic methodologies proselytized by the tech sector, such as agile and lean innovation. However, I find that the language and practices of these methodologies have become dogmatic to the point of ineffectiveness. There is much lip service to their principles and processes, but in many cases, following them has become more about ritual than meaningful outcomes—similar to how one might attend a Catholic mass, saying the prayers, standing up, sitting down, singing, and taking communion, yet leaving without absorbing the core value these rituals were meant to impart.
For this reason, I prefer a different term, one I believe I may possibly have coined: aggressive incrementalism. This approach optimizes for taking as many small steps as possible as fast as possible, with each step being as informed as possible by reality and our objectives. (As you can tell, I’m still working on a more refined definition!)
Aggressive incrementalism would tell us that the zealous, wholesale adoption of a transformative technology is unwise – that there is too much at stake to put all of your eggs into a single “adaptation basket”. At the same time, it tells us that it is OK to ask the question about AI for GI because that conversation should inform our identification of those next incremental steps.
Reverse Psy-politics
That said, there may even be a politically savvy rationale for framing the question of interconnection queue improvement around AI. The queue is a contentious issue, sparking finger-pointing and defensiveness the moment it comes up. So while a “problems-first” conversation is ideal, there’s the practical question: Will anyone actually engage in it earnestly?
Perhaps framing the discussion as “using AI to improve the generation interconnection queue”—despite the risk of hype—might be the most effective approach from a political standpoint. Could the allure of AI’s “sex appeal” be strategically leveraged to draw stakeholders into a solutions-focused conversation, with the subtle goal of uncovering core issues? It’s a bit like coaxing a child to eat their vegetables by talking up dessert.
A Multi-Agent Problem
To effectively address this question, I must consider perspectives beyond just the utility. While the utility is often seen as the source of the problem, I believe the most promising opportunities for AI-driven product innovation lie upstream with the developer and more broadly within the larger ecosystem. I’ll discuss where I see the most appropriate changes for the utility itself and identify areas where meaningful adjustments can be made within the utility without relying on AI. And as promised, those four areas of impact are captured below.
1 – Data as the Competitive Landscape
The most likely scenario I foresee for artificial intelligence—and particularly large language models (LLMs) in the near term—is that these technologies will become commodities. Current market trends, such as the open-sourcing of LLM models, suggest that this shift is already underway. As LLMs are commoditized, and as their performance and context window improve, the competitive edge will no longer lie in AI capabilities themselves but in the quality of data organizations bring to the table. In other words, the real advantage will depend on whether a utility is measuring valuable metrics and if that resulting data is accessible at a low cost.
For utilities, this means the priority should be on generating as much high-quality data as possible and curating it for multi-party use. Additionally, it’s crucial that regulators work to create an environment where utilities are safeguarded against cyber and legal risks in the process. This regulatory support will be essential to enable utilities to share data safely and to leverage it fully.
2 – Automation & Integration
AI also has the potential to improve the interconnection studies, but not with “artificial intelligence”— we’ll use our old friend “automation and integration.” While this language is no longer en vogue in today’s tech landscape, a skilled engineer knows to use the right tool for the job. There remains a wealth of low-hanging fruit that can be gathered through straightforward automation and data integration rather than complex solutions like LLMs or other AI-driven approaches. This area is one where a significant technology innovation I personally worked on has made an impact (see [ANODE paper link]).
3 – Strengthening Speculative Project Evaluations
The interconnection process involves multiple stages, some of which lie outside the formal application steps but significantly impact the overall project timeline. For instance, in the early stages, there is often considerable “experience-based” speculation aimed at valuing projects, assessing their likelihood of success, and estimating total project costs—including interconnection upgrades. This is a prime area where high-quality data and AI/ML solutions can enhance decision-making.
Imagine a comprehensive database capturing every developer project from inception through to completion (whether successful or not), documenting every aspect of its lifecycle. With such data, we could develop a predictive engine that uses early-stage inputs to estimate the likelihood of success and total integration cost. Notably, absolute accuracy isn’t as critical here as consistency. Even if we could only rank-order projects, allowing developers to prioritize capital allocation, that alone would represent a substantial improvement.
4 – Up-skilling & Improving Technical Acumen
The skills and experience required to produce high-quality, accurate interconnection requests on the developer side—and to process them on the utility, ISO, or RTO side—are in short supply. Simply put, there aren’t enough people with the necessary expertise to meet the demand. While we can specifically identify the need for more engineers with transmission or distribution planning experience, even roles in business development could benefit significantly from enhanced technical acumen. This is an insight reinforced by my wife’s experiences with her previous employer, as well as the accounts of close friends and colleagues working with various renewable developers.
This skill gap presents a ripe opportunity for an LLM-agent-based solution. A specialized LLM could be developed to understand regional market participation rules, interconnection requirements, technical standards, application processes, and even power systems engineering and other clean energy technologies. Individuals with less specialized skills could use such agents to improve the quality of their requests, thereby reducing the queue churn that leads to delays for all parties.
However, this may be a challenging area for product innovation. Given the rapid improvements in LLMs’ capabilities—such as expanding context windows and basic configurations—a sufficiently large context window could enable an LLM to access and process all relevant resources directly from the web. As a result, innovators or investors might hesitate to pursue this problem, knowing that LLM performance is advancing quickly enough to address it in the near future.
Conclusion
For the past six months, I’ve been beating myself up with the fact that I haven’t published a new post since late April. Several drafts have been near completion, yet I’ve brought none of them across the finish line. So, I was grateful when a time-sensitive assignment came my way, prompting me to finally put pen to paper on a fresh, compelling topic.
Interestingly, my last post also touched on the interconnection queue, specifically how reducing the cost of interconnection studies could lead to unintended behaviors from various stakeholders. I believe this caution applies equally to the use of AI: we must stay mindful of how different players may respond to shifts in the “rules of the game” or the costs of resources involved. Otherwise, we risk achieving the opposite of what’s intended or overlooking product innovations that truly matter. That said, in the interest of progress, sometimes the newest tool isn’t the right tool for the job. But as long as we remain honest, we owe it to ourselves to ask the question.