Work that continues

Most of what XataWorks does happens while you are watching it. Some of it does not: a prompt you scheduled, a chat another chat is allowed to prompt, a long run you walked away from. This article is what those are, and what they will never do without you.

A chat’s task, and its notes

The control in the chat header reading No task is where a conversation is given one. A task is a name plus a set of notes that belong to it, and the notes travel with the task rather than with the conversation — so work you pick up again next week arrives with what you wrote down last week.

Once a chat has a task, the menu also offers Open notes for the notepad that goes with it, and No task to step out of it again — which keeps that task’s notes rather than discarding them.

The agent can read and write those notes itself, which is the point of them: a long piece of work that runs over several sittings has somewhere to keep what it has established, instead of relying on a conversation nobody wants to re-read.

Scheduling a prompt

The chat header’s ⋮ menu carries Schedule a prompt. It sends a message into this chat, at a time you choose, exactly as if you had typed it.

The four things it asks:

• Description (optional) — what the schedule is called wherever it is listed. The prompt itself stays one click away in the chat.
• Prompt — the message, sent into this chat exactly as if you had typed it.
• Run — when.
• While nobody is watching — what to do about anything that needs approval. The default, Ask me before anything that needs approval, means the run stops and waits rather than deciding for you.

A chat that holds a schedule shows a badge in its header for as long as it does, and the badge is permanent rather than hidden in a menu. That is deliberate: a schedule spends money while nobody is watching, so a chat that holds one must never be able to look like a chat that does not. Clicking the badge lists what is scheduled, and lets you run it now, edit it or cancel it.

Letting another chat prompt this one

The same menu carries Listen for requests. A chat that is listening can be sent work by another chat — which is how a long job hands a piece of itself to a conversation that is set up for it.

A listening chat carries its own badge in the header, for the same reason the schedule does: a chat that another chat may prompt must not look like one that cannot. The badge lists what has arrived and from whom.

What happens to an approval when nobody is there

This is the part worth being sure of.

If something needs approval and there is no window to ask in, nothing is allowed. The agent is told there was nobody to ask — which it can act on sensibly, for instance by doing the part that needs no permission and reporting what it could not finish. It is not a refusal, and it is never a silent yes.

While that is happening, the chat header carries a badge saying so, with a Resume button. Pressing it puts the questions back in front of you.

That badge is shown even for a conversation running somewhere other than this machine, unlike the task and schedule badges beside it. The reason is worth stating: a schedule is local state, but a suppressed approval question is something being done to you right now, wherever the turn is running. Hiding it is the one thing this could not afford to get wrong.

What a grant would have covered, had you been there, is Approvals and permissions.

Notifications

When something is waiting for you, a bell appears in the chat header with a count on it. The count is deliberately outside the ⋮ menu rather than inside it: a count behind a click is not a count.

Checking what ran

After an unattended run, the call log is how you find out what actually happened on the sites involved — reached from the chat header’s ⋮ menu under Log, and described in Approvals and permissions. It is the one record that does not depend on reading the transcript back.