i used to think
workflows and multi-agent setups were basically just different names for the same thing since youre always just prompting an llm. turns out thinking that way leads to
massive architectural failures bc you arent accounting for how much autonomy actually changes the logic.
most people overcomplicate things by using agents when a simple script would do . has anyone found a good way to decide when to stop using a fixed sequence and let an agent take over?
link:
https://dzone.com/articles/workflows-ai-agents-multi-agent-systems