Permission gates
Also called: permission request, permission prompt, approval gate
What are permission gates?
Permission gates are the points where an AI coding agent stops before taking a particular action and waits for a person to approve or deny it. Until somebody answers, the session makes no further progress, which is why an unanswered request can idle a long run indefinitely.
The mechanics, and the cost.
The behaviour is deliberate and correct. The operational problem is entirely about where the person happens to be when it fires.
The agent stops before acting
At the point of a consequential action, the agent pauses and describes what it wants to do. Nothing further happens until it has an answer.
The block is silent
A waiting session produces no output. Neither does a session working hard on something difficult. From outside, the two are indistinguishable.
Answering takes seconds, waiting takes hours
Most requests are decisions a person could make instantly with the context in front of them. What costs time is being somewhere else when it arrives.
Why this term matters operationally.
Permission requests are where supervised autonomy either works or quietly falls apart.
It keeps a human accountable
The pause is the mechanism that keeps consequential actions attributable to a person rather than delegated wholesale to an agent.
It is the main cause of idle runs
An unattended session that hits a request stops being productive at that moment, regardless of how much work remained.
It is the case for remote oversight
If requests can reach you wherever you are, walking away from a long run stops being a gamble.
Related but distinct.
Terms that get used interchangeably and should not be.
Permission gate questions
Why do coding agents ask permission at all?
Because some actions are consequential and hard to reverse. Asking first keeps a human accountable for the decisions that matter, which is the whole basis of supervised autonomy.
What happens if nobody answers?
The session waits. It does not time out into taking the action, and it does not proceed past the request. This is safe behaviour and also the reason unattended runs stall.
Is a permission request the same as a plan decision?
No. A permission request is about a specific action the agent wants to take now. A plan decision is approval of a proposed approach before work begins. Both block progress, and Virex Remote surfaces both.
Can permission requests be answered remotely?
Yes, that is what Virex Remote is for. Approve or deny from any browser or device, and the session resumes on your own machine.
Are remote approvals recorded?
Yes. Security events are audit-logged with the actor, the target, and the timestamp, so an approval from a phone is as traceable as one at the keyboard.
Stop being the reason it is waiting.
Virex Remote brings pending requests to whatever device you have with you.