Why LLMs Are the Energy of the Future And How to Ideate for a Startup
I don’t run a startup, so maybe I’m not the best person to talk about this, but I’ve been thinking a lot about innovation and ideation lately. I wanted to understand what it means to create something meaningful, and I thought sharing my thoughts might be helpful, especially as I’m figuring this out in real time.
I’ve started to see LLMs as a kind of commodity, a form of energy with endless possibilities. Think about electricity, when it was first introduced, it transformed industries and created countless new use cases. I believe LLMs have the same potential. Sure, there’s a lot of hype right now, but I think this is more than just a trend. LLMs are here to stay, and their impact will only grow.
To break this down, I tried to simplify my thought process. Let’s imagine a company that sells water from a well. At first, the process might involve someone physically going down the well, drawing water, and filling containers. You could optimize this by hiring a highly athletic person who can do the job faster and more efficiently. You could even measure their performance by tracking how much water they draw and how quickly they do it. This is optimization within the existing process, getting better at what’s already being done and as long as your competition is doing the same thing, it doesn’t matter.
But then innovation kicks in. Someone sees an opportunity to create a manual hand pump, which reduces the effort (energy) needed and cuts labor costs. This is also a new layer of optimization. Now, within this layer, you can optimize further, maybe by designing a more efficient manual pump that moves more water per unit of energy consumed. Over time, as electricity becomes available, you can replace the hand pump with an electric motor. The motor might use more energy, but with access to a power grid, the overall cost drops even further.
I want to emphasize the idea of optimization across layers. At each stage, there’s room to improve. Hiring a highly athletic person is optimizing the existing process instead of having a slower person do the task. Introducing a hand pump is a leap forward, but even within that layer, you can optimize further, maybe by designing a more efficient manual pump that moves more water with less energy. This mindset applies to LLMs, too. There’s always room to refine and improve, whether it’s the model itself, the data it uses, or the applications built on top of it, and in the future, there may be another process that may make this seem not that efficient, driving cost to zero.
This is where I see parallels with LLMs. We now have a hardware layer built on infrastructure run by electricity and an AI layer built on hardware. The goal is to use this AI "energy" to make processes more efficient, especially by automating tasks that used to require hours of manual labour. The big question is: How are you creating value within this system?
In the B2B space, are you going vertical, focusing on a specific industry like healthcare and identifying small, repetitive processes that can be automated with AI? For example, using the vast amount of data in healthcare to streamline administrative tasks or improve diagnostics. Or are you going horizontal, building something like a CRM that can be used across multiple industries?
In the B2C space, are you creating value directly for consumers, for example, tools that help with coding, or are you working to improve the AI system itself, whether that’s by collecting higher-quality data, scaling AI infrastructure (like Scale AI does), developing vector databases, or enhancing the underlying technology? These are important questions because the way you create value will depend on the path you choose.
If you’re thinking about vertical integration in a specific industry, you’ll need to rely on domain expertise, whether it’s your own, your network’s, or through deep research into how that industry works. You’ll need to find those nuanced processes that can be automated ( which can only be known if you have experienced it yourself or through intentional research).
At the same time, it’s important not to get too distracted by the technical details of LLMs, which seem to evolve every day. You don’t need to know everything about transformer architectures or neural networks, but you should have a clear sense of their workings. When to zoom in on specific processes, and when to zoom out to see the bigger picture.
Another thing I’ve realized is that I often put too much pressure on myself during brainstorming sessions. I feel like I need to come up with an answer by the end of it, but I’ve learned that it’s better to let ideas develop over time. Not to rush it. Just explore and let things simmer in your mind.
Another thing I’d caution against is jumping into horizontal applications like CRMs, at least in my opinion. Big players like Salesforce, Zoom, or Notion will likely integrate LLM features into their platforms quickly. Similarly, competing with giants like Microsoft (think Excel) or replicating what base models like ChatGPT can already do (e.g., voice capabilities) or will do in the future might not be the best strategy since it will eat your effort. Instead, focus on areas where you can create unique value without directly competing with the big players.
One other psychological trap I usually go through is there are smarter people. It can feel really overwhelming to try and come up with a startup idea, sometimes even hopeless. But I’ve thought that maybe the key is to approach it through reverse engineering, focusing on how you can make the product happen first, even if it’s not super efficient at first, maybe even with code that takes up a lot of resources and time. Like that company using human labour to draw water. They didn’t wait around for the perfect solution, they just started with what they had. It seems like optimization can come later, but the first step is just making something exist. And once you’ve got a product that does what you want it to, I don’t think it’s worth stressing over whether there are better engineers or smarter people out there who could’ve done it "better." Maybe the point is just to start building and see where it goes.
In the end, I see LLMs as a form of energy that sparks a revolution. They can be plugged into almost any industry, and their potential is only going to grow. The application layer built on top of LLMs will create immense value, and there’s no better time to start building something of your own!