Doesn't matter how good you are at using AI or building AI agents, if you don't have the compute power or proper system design, things might feel stuck sometimes.
When I started learning AI agents, I started with LangChain and LangGraph. I used the Gemini API and many more things, but you know, the free limit used to expire.
My agents used to get stuck. As an alternative, I tried finding models to run locally on my machine.
Again, that itself was a problem because without a GPU, it was a huge problem. The speed of token generation was very slow.
It was an awesome experience for me. That time, I realised that if you don't have the compute or any paid plan for the API, you will go far, while others might not have even thought about it.
This is the first problem that a developer generally faces. But even though if you have access to tokens and good compute, still, if you don't build the proper system design, things might fall.
My first agent worked so well, but the problem was scaling. It was a bottleneck for me. Then I thought of learning about microservices and RabbitMQ. I don't even remember when I started using the system design concepts.
I felt the problem is the need for a solution, so just keep trying with your efforts.
If you are one of those who love to work with AI, make sure you learn system design.
I started learning from the Microsoft Official Website. YouTube helped me, but reading helped me to scale. For me, YouTube is my knowledge bottleneck. When you read real docs by the official authors, we don't just learn about the system; we learn about writing, thinking, reading, and many more things.