Currently, we mix agent selection with LLM selection in a single drop-down list. These two capabilities are intrinsically separate, and we should be able to run a researcher or a sales researcher or whatever against either ChatGPT or Grok or any other LLM that you add to the product. We should also probably seriously think about how we select agents. The ability to understand the capability of the agent and even enhance that agent rather than simply seeing the name in a list will become progressively more important.