Technically, it doesn’t matter. But that’s a terrible answer. The best answer is that LLMs in production are powerful and slippery.
Python coders will get best results juggling [greased bowling balls] LLMs wearing their specialized Python gloves. But when content teams borrow those same gloves, refresh rate is too low
Content Agents today work best with an automation platform like Make or Zapier,
With many more far more granular (and lengthier) LLM requests, the other thing that makes Content Agents work is automation.
As a general rule, LLMs they need three things with every request. LLMs need:
- Rules, for language and style
- Context, for SEO and audience targeting
- Examples, to reinforce 1 and/or 2
A single piece of content might need a dozen or more LLM requests, and (tuning then) providing that information with every request cannot be done manually at scale. By integrating all of the above into a single automated workflow, Content Agents make previously impossible content workflows possible.
They do require more work upfront taking on superhuman tasks (“Comb through all the news today…” ), CAs don’t just make short work of old content types, they open doors to new ones. And they unlock the scalability benefits marketers hoped we’d get when we all tried using ChatGPT to move up the SERPs.