So I'm not a developer, but I spend about 15 hours a week over last year building with agents and have developed some expert system and effective training bots. But for me Agent Zero confabulates more than any other agent platform I've worked with. Is this happening for anyone else?
It tells me there's an instance set up on the server of an application and that the testing is all fine when it isn't. It tells me it's working through a scan of selected websites and producing a report when it isn't. It often takes me a lot of prompting to get it to admit that it was simulating a web server monitoring output and it wasn't the actual output.
Experimented with a wide range of llms and still getting this problem frequently (or endless loops) including with gpt 4.1, 5.2, Kimmy K2, GLM 4.6. I'm sure it's something to do with me because a lot of very expert users are doing great with Agent zero, so appreciate any tips.