Why does this happen?
Usually because a key detail was implied instead of explicit, the teammate used general knowledge when they should have used a specific source, or you were using the same word for different things.
What if it keeps happening?
Ask Tycoon Agent to correct and re-evaluate the same Task. Your correction becomes new evidence and may update the Task's success criteria; Tycoon Agent can continue the work without creating a second conversation. If the pattern repeats, tell Tycoon Agent so she can change how future work is checked.
If the work was right but aimed at the wrong thing, see AI misunderstood my request; if it stopped partway, see AI stopped mid-task.