My experience with AI agents has been fascinating! They have proven to be incredibly helpful in various tasks, from answering questions to providing recommendations. However, it's important to be aware of their limitations and use them as tools rather than replacements for human interaction.

My experience with using them out of the box (BabyAGI, AutoGPT) is that they're extremely slow and ineffective.

But, using Agent like methods for making complex prompt sequences on specific problems can work well.

So general agents suck (for me), but using agent-like methods for very specific use cases can get decent results.

Been great. The functions LLM's write always need some tweaks and optimizations, but they're still pretty great because they show me the libraries I'll need and remove a lot of the initial friction of designing and writing.