digdatechAGI

the comment from jjmarinho about Wispr + Claude is interesting to me — voice dictation gets you into the agent, but what happens to the intent after? you've said what you want, but the system still needs to know which agent, what project context, what constraints. the dictated note is still unstructured at that point.

curious whether mercury has any primitives for capturing that "pre-dispatch" context or if the assumption is that the human has already structured the task before it enters the no-code layer.

one thing i'm watching is whether managed agents changes the 'first instruction' problem. right now, even with great orchestration, someone has to decide what the agents work on — drafting that initial task spec, choosing which agent gets it, providing the right context.

that step is still manual in every setup i've seen. curious if managed agents has any primitives for that or if it's out of scope.