I was watching this video about a developer who was skeptical about LLM-assisted coding, but gave it a real shot over a month. He doesn’t plan on completely giving into the “vibes” or anything, but his opinion of how this sort of development works and what it can and can’t do have evolved quite a bit from his first impressions.

For the record, I think LLMs are pretty good for a bunch of things, but it wasn’t until I started using them for code in December of last year when I truly had my mind blown. The tools have only gotten better since then.

I wanted to reference this video because a lot of the developers I've seen on social media who complain about LLM coding being terrible tend to imply they've spent 10 minutes using something like Cursor or Claude Code, it didn't do exactly what they wanted, and they took to social media to complain. Imagine if I opened up VS Code, used it for 10 minutes to try and write a Python script, and declared it useless because I couldn't get a working script going in that time. No, I haven't taken any time to learn Python or macOS automation, why do you ask?

I bring this up because I genuinely think there's something meaningful here with LLM-powered engineering, and I think that just like add engineering tools, you need to learn how to use it before you can effectively use it or provide insightful commentary on its merits.