There is no doubt that Agentforce for Developers is incredibly useful to write Apex, and test case generation is great!
However, I find it easier to actually write custom code with ChatGPT or Claude than to understand how to get things done on top of Salesforce solutions. Today, it's easier to write custom code in React using libraries, without spending months studying each library, than building on top of Salesforce. The advantage of building WYSIWYG in the cloud is now being eroded by the idea-to-code approach that LLMs have introduced, where developers get much more productive.
If I want to develop a solution on top of - say - Revenue Cloud or Data Cloud, or CPQ I need an LLM to tell me how to get things done. The help.salesforce Agentforce is much worse than ChatGPT or Claude, not even pointing to the right documentation that would answer my prompts, while ChatGPT or Claude don't understand enough about how Salesforce works.
Apologies to the owners of this group, this goes beyond Agentforce for Developers, yet improving Agentforce overall to really leverage LLMs for Salesforce end-to-end development is a must.
