An AI employee that appears idle may be between scheduled shifts, waiting for a required sign-in, or showing the result of a failed shift. Those situations call for different next steps. Start with the status and its reason, then use the Journal or Inbox to find evidence before changing the goal or starting work again.

This guide was prepared with AI writing assistance and checked against Orke’s public AI employee documentation. BOMTOBE makes Orke. It describes documented behavior, not a support case or a test of a live employee.

Read the status and the reason together

The line below an employee’s name gives a reason, such as the next shift time, and may include today’s recorded action counts. “Working,” “Resting,” and “Stopped” describe different situations, so read the full message instead of treating the short status as a diagnosis.

Use this order: read the current status and its reason; open the relevant shift card in the Journal; expand “Actual actions” when available; check Memory or Inbox if the reason concerns the plan or an instruction; then take the smallest matching step and check the status again.

The Journal contains the employee’s account and Orke’s browser action records. These are different kinds of evidence. For an important result, check the destination itself as well as the Journal before concluding that the result is available.

When work cannot start or is resting

The documentation says “Why work cannot start” appears when an employee cannot begin. Common reasons include a missing goal or a required sign-in for the working AI. If the goal is missing, provide the assignment in a meeting. If the interface asks for sign-in, sign in to that AI again, then choose “Work now.” Changing the goal will not resolve an account sign-in requirement.

Orke must be running for employees to work. Employees resume when Orke starts again, and missed shifts are caught up only once. If the PC was off or Orke was closed, check that it is running before treating the schedule as broken.

“Resting” can show when the next scheduled shift will begin. If that time falls outside work hours, it moves to the next work-hours opening. This is a schedule explanation, not evidence that the last shift failed.

The “Nothing to do · Next shift …” pattern means shifts have repeatedly finished quickly without work, so Orke lengthened the rest period, up to 30 minutes. Check the Memory tab’s status file and Journal to see why the employee thinks there is nothing to do. An instruction clears this delay. Use “Work now” only when you want an immediate shift.

If a shift failed or was interrupted

Open the specific shift card in the Journal. Orke lists “Failed,” “Interrupted,” and “Stopped” shift statuses. It also describes “Repeatedly waiting to retry”: check the failed shift’s Journal reason; if it is a usage limit, work resumes when the limit clears.

Do not infer a usage limit from a quiet screen or a retry alone. Read the Journal reason. If it names another failure, follow that evidence instead of switching accounts or rewriting the goal speculatively. Before encouraging another attempt, check what the previous shift did and whether it created an external side effect.

When “Actual actions” is available, inspect those records alongside the employee’s summary. Orke records browser actions such as opening a page, typing, clicking, posting, and closing it. These records help distinguish a claimed step from an action Orke observed; an important destination still needs to be checked.

If sign-in, setup, or instructions block work

For a required working-AI sign-in, sign in to that AI again and choose “Work now.” For a missing goal, provide one in a meeting. If the status is “Setup error,” follow the file and line named in the reason; if you edited employee.md directly, check its format. After a change, read the new reason if the status does not change.

Instructions are sent through the Inbox. The notice below the input field explains when the employee will read one. If work has not started, start it so the instruction can be read. Use “Send and work now” to have it read immediately. Check that notice before sending the same instruction again; duplicating an instruction does not make it active sooner.

If a human request appears again

Requests are for steps only a person can do, such as entering a password, completing verification, paying, solving a CAPTCHA, or using an account that requires the person’s action. The employee can continue other work while a request is pending and will not create another pending request. When the person completes the requested action, verify the result in the relevant account or browser window, then resolve the request as “Done.” Choose “Won’t do” if it will not be completed; “Give my opinion” is available when direction is needed. A resolution records the choice; it does not perform the human-only action itself.

Keep a short diagnosis record

For a recurring issue, record the exact status and reason, the relevant Journal entry, the action taken, and the next observed state. Orke’s employee memory separates status, plan, lessons, and dated journal files; use the place that fits the workflow. Do not copy passwords, verification codes, or other secrets into notes.

BOMTOBE makes Orke, a desktop workspace for AI agents including Claude Code and Codex. Read the current AI employee documentation: https://orke.bomtobe.com/en/docs/orke/employees/