How Can Builders Probe the Machinery of AI?
7 categories of AI opportunities
Welcome to Infinite Curiosity, a newsletter that explores the intersection of Artificial Intelligence and Startups. Tech enthusiasts across 200 countries have been reading what I write. Subscribe to this newsletter for free to directly receive it in your inbox:
Someone recently asked me how I think about the landscape of AI as an investor. As Kyle Harrison likes to say, I decided to write it down so I can read what I actually think.
Understand the machinery of AI help us look for opportunity. The way I like to frame this is:
AI is the goal.
ML is a vehicle to get there.
DATA is the fuel for that vehicle.
The machinery of AI is delightfully complicated, which means there are lots of opportunities for founders. In this post, I’m proposing a framework to see how we can probe this.
Category #1: Paving the paths to different destinations
This refers to AI software that helps people do different pieces of work. As a startup, you need to identify a target and pave the path for people to get there. They are usually agnostic to the type of vehicle being used. And they use different vehicles depending on the needs of the customers.
If people like it, they will pay you to use the path to get to that destination.
AI software has come a long way, but it’s still so early. The models we use today are great at generating text, but they struggle with the complexity of the real world —long-term memory, reasoning, autonomy, and more. AI isn’t just about answering questions. We need AI to handle complex workflows, improve over time, and integrate seamlessly into daily life.
Where do we go from here? One promising direction is self-improving agents that don’t just generate outputs but also learn from feedback and adjust accordingly. Another is memory-augmented AI where models that can recall context beyond a single session, making them far more useful for real-world tasks. The companies that solve these challenges will unlock entirely new applications beyond chatbots and copilots.
Category #2: Building the vehicles and renting them out
This refers to startups building AI tooling for companies to use. They want companies in category #1 to use their vehicles and pay rent. If more companies use your vehicle, you’ll earn more revenue.
You can’t have a single vehicle to serve every use case in the world. There’s a reason we have cars, trucks, motorbikes, minivans, and so on. They serve different purposes and they all have customers. The goal is to figure out what you’re good at and then build that vehicle.
Category #3: Supplying the fuel
This refers to startups supplying data in different shapes and forms to AI companies. For example, this can be sensor data for automotive companies. Or it can be a large database of buyers with names/emails in a specific sector. Or it can be labeled images for specific tasks.
Category #4: Developing new ways to process the fuel
This refers to startups building processing engines to process the data in different ways. It can be companies providing data labeling products to different sectors. Or it can be companies providing ETL products. Or companies providing vector databases. Or companies providing data warehouses.
Category #5: Building the physical body to let the vehicle interact with the real world
This refers to Physical AI. These are AI systems that can learn from and interact with the physical world. Robotics is a good example of this. AI is great at working with digital content, but it struggles in the real world. Physical AI requires a good amount of human intervention to function well. How do we get to fully autonomous systems that can navigate dynamic environments, manipulate objects, and perform dexterous tasks?
The key challenge is data. AI models need to be trained on real-world physics at scale. This requires better sensor fusion, reinforcement learning in real environments, and architectures designed for physical interaction.
The opportunity is massive! Whoever builds the AI that can truly “see” and “understand” the physical world will define the next industrial revolution.
Category #6: Enabling builders to build
The AI infrastructure boom is in full swing. But where are the real pain points? Compute costs remain high and inference is still too slow. Most AI companies are limited not by ideas, but by access to affordable and efficient processing power.
The biggest opportunities lie in optimizing inference, reducing latency, and making AI development more efficient. Companies working on low-cost model serving, more efficient architectures, and alternative compute paradigms could reshape the entire landscape.
As AI adoption grows, the demand for cost-effective infrastructure will only increase. More builders would want to participate in this movement. And we need accessible infrastructure to enable that.
Category #7: Enabling the vehicle to transact
Today, AI can generate content. But it can’t act in meaningful ways. It can write an email but not close a deal. It can summarize a contract but not negotiate terms.
The next frontier? AI that moves money, executes transactions, and autonomously interacts with digital markets.
A real AI economy means autonomous agents that can buy, sell, negotiate, and execute on behalf of users. It also means building the necessary trust and safeguards, including AI-native insurance products to manage risk when autonomous systems take action. The companies that enable AI to participate in the economy will create entirely new markets.
What’s next?
With AI, we always need to keep asking ourselves—What’s possible next? How do we build AI that learns, remembers, and acts with real-world intelligence? How do we enable truly autonomous systems that create real economic value? The most exciting opportunities in AI lie in bridging these gaps.
The future belongs to the founders who ask the right questions and push AI beyond its current limits. If that’s you, there’s never been a better time to build.
If you're a founder or an investor who has been thinking about this, I'd love to hear from you.
If you are getting value from this newsletter, consider subscribing for free and sharing it with 1 friend who’s curious about AI: