This video by Theo is excellent. I watched it about a week ago and have been mulling it over all week. It finally solidified all those thoughts that I was trying to formulate from my own experience.
A significant part of the video has interesting reasoning about defining the word "vibe coding" and how everyone understands it differently. He establishes that agentic coding is when you use AI agents, and vibe coding is its subset when you don't really read the code in the end. Couldn't say it better, it seems. But really it's a spectrum, of course, it's important how much you don't read.
Later he throws out theses justifying vibe coding and explaining how to use it, which I'll freely quote:
- If you want to be a coder, then understanding how code works is necessary
- There's a lot of code that's useful to have, but not worth writing and reading
- Vibe coding solves needs that are currently solved by countless micro-libraries. Most people just want a solution to their specific small problem, and both give them that.
- Vibe-code is just a subset of legacy code. Both are code we forgot about/code we don't understand.
- If you understand that agents write code better than you - don't use them. Agents should write more and faster, but not better.
- If you don't see applications for vibe coding, then you don't have much imagination
- Vibe coding is good because throwing out someone else's code isn't as painful as throwing out your own
And I agree with all of this one hundred percent, even though they're kind of about different things. Agents are wonderful when you know what to do, can explain it, and can double-check. Agents are so-so when you're trying to do something that's outside your area of competence.
If you don't know what needs to be done to complete the task - take away the agent's write access, let it only read and suggest. And you'll learn from that. If you know - let it do it, and you check. If you don't understand anything at all, and hope it will do something good - you're fucked, to quote less freely.
And at the end there's a really inspiring topic. We live in a world where you can vibe-code software that will solve your specific problem. Yes, it won't be a production solution, it'll be crooked, lopsided, but it will solve it. That is, writing micro-applications for yourself alone is a new reality. Before, you wouldn't start many things simply because the development costs exceeded its usefulness. Now this line has clearly shifted.
Highly recommend watching the whole thing. Each next phrase is such that I wanted to think about them. And the comments are gold.