How denials train the agent
Your home screen is the queue, and it sorts the agent's work so nothing has to be hunted for. "Needs you" holds actions waiting on your decision. "Happening now" shows automatic actions still inside their safety window, each with a live countdown. Separate groups cover what has already been handled and what is coming up.
An item lands in "Needs you" for a clear reason — it is above a threshold you set, it is one of the actions that always requires a human, or you simply have not enabled autonomy for that feature yet. Everything the agent handled on its own moves to the handled group with a plain-language note of what it did and why. When you are caught up, the queue says so; new decisions appear the moment the agent needs one.
Each item in "Needs you" has inline Approve and Deny buttons. Approving lets the action proceed. Denying asks you for a reason, and that reason is passed along — emailed to the tenant where a tenant is involved — so a decline is never silent or unexplained.
Higher-stakes money actions ask for an extra confirmation rather than a single tap, so an accidental approval is hard to fire off. Whatever you decide is written to the audit trail with a timestamp and your name attached.
Open the app to your home screen and find the "Needs you" group.
Tap an item to see the full context — what the agent proposes, the amount, and its reasoning.
Choose Approve to let it run, or Deny to stop it.
If denying, enter a short reason; it is recorded and shared with anyone affected.
How denials train the agent
Your approvals and denials are not throwaway taps — they are the agent's feedback loop. Every decision, with its reason, is recorded against the kind of action it was. Over time that history is what the help agent uses to suggest widening (or tightening) a threshold, always with the evidence behind the suggestion, so you can see why before you agree.
The point is that the agent earns more room by getting your calls right, not by assuming it. You stay above the loop: the agent proposes changes to its own limits, and you decide whether to grant them.
There are two ways to walk something back. Cancel stops a pending action before it ever executes — available while it is still inside its soft-commit window in "Happening now." Undo reverses an action that has already committed, for the reversible ones. Both sit on the agent activity feed alongside the "why" behind every decision, so the record of what was proposed, what you decided, and what actually happened stays complete.