What I’ve Learned About Gen AI Building Blocks

I’ve been spending more time understanding what actually happens behind the AI tools I use, and one thing I’ve realized is that using AI effectively is not just about writing better prompts.

Here are a few things that stood out to me:

1. Understanding How LLMs Work

Earlier, I mostly thought of an LLM as a tool where I give a prompt and get an answer. Learning about tokens, context windows, function calling, and structured outputs changed that perspective.

For example, understanding context windows helped me realize why an AI model may lose track of information in longer conversations. Structured outputs and function calling also showed me how LLMs can interact with applications instead of simply generating text.

2. Understanding the Modern AI Stack

The AI ecosystem is much bigger than just ChatGPT or another single model.

There are model providers, AI tools, platforms, agent skills, and even vibe-coding workflows. I’ve found that different tools make sense for different tasks, and choosing the right combination can sometimes matter more than simply choosing the most powerful model.

3. The Practical Side of AI

One thing that is easy to ignore when experimenting with AI is the practical side.

AI applications have to deal with latency, API costs, reliability, and performance. A model might give excellent results, but if every request is expensive or takes too long, it becomes difficult to use at scale.

My biggest takeaway so far is that building with AI is not just about getting an impressive output. It’s about understanding the technology well enough to make the right trade-offs.

Still learning and experimenting with these concepts, but they’ve definitely changed how I look at Gen AI tools.