As a Senior Engineer, Here’s How I Actually Use LLMs to Program

Large language models (LLMs) like ChatGPT, Gemini, and Claude are shaking up how we write software, but let’s be clear: they aren’t magic wands.

I hear this a lot:
“Can’t I just tell the AI to build my app?”
“Why not automate everything?”

Here’s the truth: as a senior engineer, I’ve learned when to lean on LLMs and when to back off.

Let me break it down with three mental models I follow every day.

I know what I know

I already know the domain, the rules, and the architecture. But do I want to spend three hours typing boilerplate? No thanks.

Example:
I need a REST API in TypeScript for a known schema.
I hand the LLM:

“Here’s the user schema. Here are the validation rules. Generate the Express controller, service, and unit tests.”

Now I can scan the output, because I know what good code looks like.

Here, the LLM is an accelerator, not a replacement. It saves me time, but it doesn’t replace my expertise.

I know what I don’t know

I don’t know Rust, but I understand core concepts, such as memory safety, types, and lifetimes.
So I ask:

“Show me how to build a simple CLI in Rust that parses arguments.”

The LLM gives me a starting point. But I don’t blindly trust it, I debug, read the docs, and iterate alongside it.

The LLM becomes a pairing partner, helping me stretch into adjacent knowledge. It’s great when I have enough context to guide and sanity-check the journey.

I don’t know what I don’t know

If you ask an LLM to build something in a domain you have zero understanding of, you won’t catch the nonsense.

Example:

“Write me a full-featured blockchain app in Solidity with security best practices.”

If you don’t understand Solidity or smart contract security, you won’t spot the subtle bugs, or the glaring ones.

LLMs confidently hallucinate. They sound right, but can deliver brittle, incomplete, or even dangerous solutions.

Here, you need to hit the books — learn the foundations yourself before you ever loop in the LLM.

Final Takeaway

LLMs are like power tools.

✅ Use them to speed up what you already know.
✅ Use them to explore what you partially know.
⚠️ Don’t use them to fake expertise you don’t have.

When you combine human judgment with AI acceleration, that’s when the magic happens.