When Things Go Wrong With AI-Generated Code

My first bad experience: the code generated by Claude Code made my dashboard unresponsive in my browser. Eventually, the data stopped updating. After a ten-minute debugging session, I asked Claude Code to revert the change, and it did so promptly. But then I started getting execution failure notices on Discord. A lot of notifications. Then I started investigating…

It appears the browser was making frequent refresh requests to one of my workflows, which depleted my Claude pay-per-use credits. Bummber. Looking at my n8n dashboard, I saw that one of my workflows was failing because of that. Logs were confirming the problem with the interaction with Claude AI. As shown on the graphs below, my instance CPU usage went through the roof. Ouch. Now I know what happened, and the problem was fixed. Now, I should find a way to rate-limit this type of behaviour. That’s for tomorrow, I guess. ๐Ÿ˜…

One of the frustrating aspects of LLMs is their lack of consistency unless you develop specific skills, which can take time to implement effectively. For example, I wanted to generate documentation for my most recent n8n automation workflow, but Claude was unable to do it, and I can’t remember the prompt that finally made it possible. I should have saved it somewhere for easy retrieval. I’m wasting precious credits. ๐Ÿคฆ๐Ÿปโ€โ™‚๏ธ

Apparently, people are barely using Stack Overflow to ask questions, thanks to LLMs and AI. I expect a similar trend among people in a community like this one on Micro.blog. Some questions would be super easy to answer by asking ChatGPT or the like. I do understand that many people still want this human touch, though.

Updated my n8n instance from v2.0.3 to v2.2.4. Super easy to do (I’m using the Docker Compose installation provided by the DigitalOcean 1-click install droplet. Took a droplet snapshot before, just in case something goes wrong. So far, so good. Of course, Claude helped me out on this. I’m not a Linux or Docker expert. ๐Ÿ˜