>> I'm curious, is anyone using any kind of LLM agent workflow in a client facing production environment?

Yes. Not everyone will admit as most big companies have policies against it. Small companies are, because the cost saving is so high.

My employer is. The LLM agent did not replace any humans, though. It was created as a new value-add service for clients.

Yes, recently upscaled research for my iOS app. Have made a pipeline for fact extracts from various sources, including some sentiment analysis. Would have taken me years to do it manually. Will mainly be value adding for end user . Not directly client facing since this is done offline and shipped as data into the final product.

Yes, in public safety to make response suggestions to emergency operators in time critical situations when texting is the mode of communication.

This seems... like a terrible idea on the surface. The cost of LLM hallucination in a time critical crisis situation seems immense

Yes. We've launched multiple production instances of this across a handful of clients. In general humans using the new workflows show very significant performance and quality improvements over unassisted humans so far for the well-defined tasks the workflows are tailored to perform.

can you elaborate?

Probably not without violating an NDA. What would you like to know?

AWWWW... It would be great to share some "meta-data" on the usecase - i wonder whether it is narrow task (that can be done via CRON job), or something that are more general and can be creative and not in control sometimes. What kind of tools that the agent use? (Web-search/ Generating content/...) Did you made any modification to the code or use it "as is". if so- how much time did you spend on the modification and in what kind of areas...

Feel free to answer only subset of the questions that you feel comfortable with them.

Yes, the key is to implement clear roles, responsibilities, relationships, and communication channels, as the OP article explains.

allofus.ai

our entire focus is on multi-agent workflows based and trained on real peoples published work. I think the output is amazing, but I'm biased of course. Its ability to accomplish soft tasks like advice, help etc is far superior to vanilla GPT4