AI, the double edged sword of Early Career Engineers

I’m going to talk about how LLMs have both benefited the beginning of my career and simultaneously hindered it.

I was probably one year into my new grad job, and things were moving pretty slowly. Of course, that’s expected at a big company, but at the time, I felt like if I kept going at that pace, I’d never grow. So I started building side projects around this time.

While I was learning new tools like Next.js, Hono, Bun, etc., I was also using Cursor. What a mistake that was.

The autocomplete was such a new source of dopamine, and the prompt engineering aspect was just so easy that I thought to myself, why bother coding most of this by hand? Mind you, this was around the time when LLMs were nowhere near as good as they are as of writing this, but I saw the vision. I saw the potential that we would eventually get to the point of not writing any code at all.

Needed a feature? Use AI.

Needed a bug fixed? Use AI.

Tests written? Use AI.

Don’t know why you’re getting an error? U S E A I.

You see where I’m going with this?

After building multiple Next.js projects, I thought I was a pro. I thought I could definitely work at a startup as a Senior Front End Engineer.

Oh boy, was I wrong.

I interviewed with a startup in stealth mode that was looking for a Front End Engineer, and I completely shat the bed. When I stumbled upon an error, I froze and didn’t know how to handle it. Some things weren’t rendering correctly on the client side, Tailwind wasn’t handling conditional colors well, etc.

From that experience, I learned a very important lesson: you gotta know your shit.

From that point onward, I made the executive decision to not use AI to write all my code for me, but instead to help me learn. For example, whenever an error appears, I don’t just ask AI to fix it for me. I ask what the error means without giving it any context. Also, I started using Neovim lol.

As I became less “anti-AI,” I slowly started running agents to handle redundant tasks like logging, tests, comments, etc. This is when I created my own personal rule of thumb:

If there was a SEV call at 2 a.m. and there was an issue with my code changes, would I be able to fix it?

If the answer is no, do it by hand. If the answer is yes, let the agent do it for you.

As simple as that.