Mostly interested in how agents should be allowed into real workflows, not just how they execute once inside them.
submissions | comments