How it is able to have a complex task: 1. To have ability to craft an idea for a blog post without a context. Exp: which topic to write for, what will be the outline of the blog post, what are the hotlines for today for the niche, what are the relevant keywords, what are the competitors on the market, is your topic hot enough to grab the users attention, is your blog topic is with good enough click-attention context. 2. What about delivery of this blog post - to which users you are going to share the new blog post - to all or only to those that shows intention to read for that specific topic? 3. Are you going to market this blog post or not, Where to market it, To which audience? All that questions could come to the mind of a person or a team, but never to a single LLM. That is why I am researching for a mechanism to make this llms with more context and ability to take actions. Of cours you can always open your chatgpt and prompt it: Write me a blog post.

@Jordy Backes that sounds awesome! Do you output each agent the output or the entire crew output? Also do you have something as a crew's manager who is keeping the crews aligned to keep the target?

That is great content @Brandon Hancock! Congratulations for your efforts! It is really helpful and a lot of handy insights how to use the crewai! I love it! I stumble on one wired thing. Because I am using gpt-4-1125-preview and this LLM actually has abbility to be forced to output in JSON when it is interacting with tools that expects json format it starts with for example: Action Input: ```json { "site_id": 402, "primary_filter": "category", "days": 90, "sort_by": "units" } ``` instead of Action Input: { "site_id": 402, "primary_filter": "category", "days": 90, "sort_by": "units" } The question is if you have a workaround of this behaviour? I did not find anyhow to tell the LLM to use this parameter: https://platform.openai.com/docs/guides/text-generation/json-mode