Vibe coding with Grok

I've been making a lot of vibe coded scripts recently to accomplish little personal productivity tasks. If you didn't already know, vibe coding is where you explain to AI what it is you're trying to achieve, and the AI spits out code which attempts to match your requirements.

I've been using ChatGPT to do these little tasks for me for a while, and it's relatively good at getting the basics down. It does start suffering from issues after just a few refinements however, where it'll introduce bugs that weren't there before, and then never be able to fix them with any amount of prompting. In this way, it's good to get a very rough outline of what you're trying to achieve, but it can't get you all of the way there.

I should note at this point that the code output it gives you isn't always to a good standard, and often it will go about things in a convoluted way – almost like it's lifted a solution off of Stack Overflow, which was an answer given by somebody with limited experience several years ago. At this point I believe it still requires a skilled developer to understand and correct these problems that AI throws at us.

The other day then, I decided to try and semi-automate a Kickstarter I've been tracking; the eufyMake E1 UV printer. I had made a little Excel spreadsheet for myself to keep tabs on how many of each reward had been backed, and then calculated from that how many add-ons people were purchasing on average, but as you can imagine this was a time consuming process and required me to update the cells myself every time I wanted an update.

Roll in Grok, whom I decided to test out. I told Grok what it was I was trying to achieve and explained the premise of my spreadsheet, so imagine my surprise when Grok nailed what I'd asked for in its first attempt. I kept asking for refinements, and it kept delivering them. Sometimes there were issues, but I outlined what they were and Grok thoughtfully analysed them, then told me what I'd asked for, why it wasn't working, and how it was going to achieve a fix – in this way it seemed to be much more capable than ChatGPT.

It's fair to say that Grok delivered 80% of the end product I have now in about 20% of the time I spent on it. It managed to deliver continuous improvements quickly and consistently, and certainly much more capably than ChatGPT, but it wasn't without issue. Occasionally it would throw in a random word in its code output, causing an error. Sometimes it would produce an incomplete line of code and transition into a thought instead – once that thought was even in French!

I feel confident at this point using Grok as a tool to aid my development but it is important to remember this: it is only a tool. It isn't yet ready to completely replace developers, nor is it a reliable tool for teaching developers the best / semantic / accessible way to do things. But I look at the state of the web in 2025 and I do think to myself that perhaps we're not very good at teaching ourselves those things the regular way either.

You can check out my Kickstarter script, and let me know what you think.