The AI Can Now Ask You a Question — and Wait for Your Answer

July 11, 2026 · InfraScout Team

Every assessment eventually hits a fork the AI cannot resolve on its own. Three hosts match the name you gave it. A configuration is unusual but might be deliberate. A finding may or may not be in scope for the review you asked for.

The AI could always ask you about a fork like that — it wrote the question into its answer, ended its turn, and waited for you to type a reply. That worked, but it made asking expensive: the turn stopped, and getting back to work meant composing a response in prose and letting the AI pick the thread back up on the next turn. Asking a question you could resolve in two words cost more than it should have.

Now it asks with a card, you answer by clicking, and the same turn carries on from where it paused.

Click an answer, and it keeps going

When the AI needs a decision from you, it renders an interactive question card in the chat and pauses on it. The card can carry up to four questions at once — so a single pause can settle a cluster of related unknowns rather than interrupting you four separate times — and each question offers clickable options, plus a free-text box for the case where none of the options is quite right. Most of the time you will not type anything at all: you click the host you meant, or the scope you wanted, and that is the whole interaction.

What follows is the other half of the improvement. Your answer goes straight back into the turn the AI was already running, so it resumes from exactly where it paused, with everything it had already gathered still in hand. There is no new turn and no re-prompting — the work that led up to the question is still there underneath it.

Together those two things change when the AI is willing to ask. A question that costs you one click and costs the assessment nothing is worth asking at every fork where a wrong guess would be expensive: which of three hosts you actually meant, whether a finding is in scope, whether a non-standard configuration is intentional or a problem, or whether an assumption holds before the AI commits twenty minutes to a sweep built on top of it.

Waiting, timing out, and moving on

While a question is pending, the conversation shows a clear waiting state, so a paused assessment never reads as a hung one. You can see at a glance that the AI is not stuck or broken — it is holding, because it asked you something and has not heard back.

It will not hold forever. An unanswered question times out on its own, after ten minutes by default, and the AI then decides how to proceed. A timeout is not an error and it does not fail the run: it is information, and the AI treats it as such — usually by falling back to the most defensible assumption and telling you which one it picked. You can also dismiss a question outright if you would rather it carried on without your input. Either way, the assessment keeps moving.

Answer in the window that asked

A pending question lives in the conversation view where it was raised. If you reload the page, or open the same conversation in a second window while a question is waiting, the card renders as expired and you cannot answer it there. This is worth knowing up front: if you see a question, answer it in the window that asked, and avoid refreshing while one is open.

The other half of a real conversation

This is the natural companion to mid-loop steering, which let you interrupt the AI while it works. That gave you a voice in a running turn. This gives the AI one: now it can interrupt you. Between the two, an InfraScout chat finally behaves like a conversation in both directions — either side can speak up mid-run without tearing down what is already underway.

It also fits alongside the confirm-before-changing behavior, though the two are doing different jobs. A confirmation is a gate on an action the AI has already decided to take, and your answer is yes or no. A question is asked before the AI knows what to do, and your answer shapes the plan rather than approving it. One protects you from an unwanted change; the other keeps the AI from working off a wrong assumption in the first place.

Questions are for live chat only, and that is deliberate. An Agentic Workflow or a task agent runs unattended — that is the entire point of scheduling one — so there is nobody at the keyboard to ask. Rather than have those runs stall on a question no one will ever see, InfraScout simply never offers them the ability to ask, and they proceed the way they always have.

Try it

Give the AI something genuinely ambiguous and watch what it does with it. Ask it to review "the domain controller" when you have several, or to assess a host you know carries a deliberate non-standard configuration. Instead of picking one and hoping, it should stop, show you a card, and ask which you meant — and once you click, carry straight on from where it paused.

See the chat guide for more on running assessment conversations. Questions or feedback? Reach us at info@infrascout.cloud.