Hi everyone. I'm still pretty new to DevOps and was exploring CrewAI for a potential side project. I wanted to automate a simple research task: scrape a few URLs, summarize, then draft a report.
My first instinct was to just script it with the OpenAI API, making a few calls in sequence. Then I tried CrewAI, which abstracts that all away. The difference in cost was... wow.
My manual script made three separate GPT-4 calls. CrewAI, for the same output quality, ended up making *way* more internal calls between agents (like a manager agent reviewing each step). My $10 test ran out in minutes with CrewAI, while the manual script cost pennies.
I love the idea of frameworks like CrewAI and LangChain, but I'm starting to think they add a lot of overhead for simple workflows. Has anyone else run into this? For those using CrewAI in production, do you just accept the cost or is there a way to tune it that I'm missing? Maybe I'm using it wrong 😅