If you work in tech and you’re not vibe coding, you’re stuck in the past, my friend. Let me explain why right away.
It’s something we tend to resist, but it has practically already become part of our daily routine, especially for small, one-shot applications: the kind you build once, use, and then forget about. Here’s a good example.
The job
I had some videos uploaded to a private software-as-a-service platform, and I had to take them, move them to my own server and build a small interface to serve them. There were several obstacles.
The first was extracting the course structure from that platform. Locally, on my machine, I only had the list of MP4 files, and I somehow had to match the structure of courses, modules and lessons with the MP4 file names, which were slightly different - the kind of thing that can drive you crazy. Then I had to build the little frontend.
I vibe coded all three pieces, very happily. It took me two or three hours in total. If I had written the pieces by hand - and I knew how to do all of them - it would still have taken me a couple of days, one thing and another.
Step 1: extracting the structure
To extract the data structure of courses, modules and lessons from the platform, I would have had to either run a little script in the browser console, hacker style, or download the HTML and parse it with Python, something like that.
Instead, I solved it by printing the page to PDF, loading it into an LLM and telling it: make me a JSON in this format for this PDF. Boom, boom, boom: I had the JSON, the data structure for each course.
Step 2: matching lessons to files
Then I moved to Visual Studio Code, where I have Roo Code, a VS Code extension - but there’s Cursor, Cline, Void, there are three thousand of them.
With a bit of hammering, I had Roo Code analyze the JSON with the lesson names and match them with the names of the MP4 files on my disk, then update the JSON structure so that each lesson pointed to its MP4. It also checked for duplicates, empty entries and mismatches - MP4s mentioned in the JSON but not on disk, and vice versa. And it did it, and did it well, after some hammering. I ended up with JSON files containing not only the course structure but also the MP4 for each lesson.
Step 3: the frontend
Then I had it build the little frontend. It came out buggy as hell, but it did it in five minutes, and in about half an hour I fixed the bugs. At my request it wrote it in vanilla JavaScript, and it’s a simple little frontend where you browse, click and watch the videos again. I may need to add authentication and a few more complicated things, but especially in one-shot cases like this, in my opinion there’s no going back.
Stop resisting
There’s no such thing as a programmer who doesn’t work alongside these tools anymore. If you don’t, you’re stuck in prehistory. Stop resisting, folks.
To wrap up: an experienced programmer with a language model or a coding agent at their side is a beast. A junior programmer with this assistant in hand learns faster, as long as they don’t hand it too much responsibility - that’s the delicate part, in my opinion. And if you don’t program at all, you can learn to program, because you have it do things and explain them to you.
Computing is becoming something different. I’m happy about it, because it brings us closer to the people who do design and product. There have always been lots of misunderstandings caused by the different iteration cycles on our artifacts. A designer iterating on a prototype, mockup or wireframe works in windows of two hours, three hours, half a day, a day. A programmer iterating on a piece of code that produces an interface typically needs longer, so they become more resistant to changes and proposals - and they have a point - and these people don’t understand each other. In my opinion, now they do. That’s a great thing.