Yes. I have tried various agents over the last ~1.5 years on multiple occasions on a bunch of different kinds of engineering type tasks. So far there has been a total of 1 time where the output was reasonable enough that I could build on it and not feel ashamed of the result (and that time probably saved me like half an hour). All other times, I wasted a bunch of time debugging crap and then just wrote the thing from scratch myself.

The closest I've come to somewhat consistent success with them is when I struggled to come up with a good search query for an issue I was having and after asking a longer prompt to an LLM it either gave me a close enough answer that I could figure it out from there, or the answer included some keywords that helped me come up with a query that got the results I needed.

By and large, I consider them crap for anything beyond the basics. On the other hand, I absolutely understand why they may look great in cases where the person using them doesn't have an idea of what the output should look like. They're a minimal productivity boost at best, at an insane cost.