Absolutely. So in the previous version you could spin up multiple agents - that each ran a prompt. So for example, you could have an agent named Mark that ran a prompt which had context for your codebase, whereas you could have another agent Amy that had, say, a prompt that would help Amy give you writing feedback.

When you logged into the app then, you'd have to choose the agent to chat with - in the same way you'd choose a teammate to chat with in Slack - based on their area of expertise.

This system works well if you only have a few agents, but once you reach a certain number, it becomes a hassle to determine who to chat with.

Ideally, the user shouldn't care about which agent would response / which prompt is applied, they should just care about what they want, and the system should be smart enough to apply the right context / instructions accordingly.

And this realization led to the redesign.