> ## Documentation Index
> Fetch the complete documentation index at: https://docs.cogniagent.ai/llms.txt
> Use this file to discover all available pages before exploring further.
# Keep working
> Tell your AI coworker to keep going until the job is genuinely done, instead of stopping to check in after every step.
For a job with a clear finish line, tell your coworker to keep going until it's actually done.
## What it does
Left to itself, a coworker tends to do a chunk of work and hand it back. That's right for a question and wrong for "go through all 200 of these".
Type `/loop` and it keeps going: finishing a piece, checking what's left, starting the next, until the job is genuinely complete. You can add the task on the same line, or an interval like `10m` to have it check back on that rhythm.
While it's running you get a marker showing it's still going and how many rounds it has done, with Stop right there.
Type `/stop` to end it.
## When you'd use it
* A long list of items that all need the same treatment
* A job you know will take an hour and don't want to shepherd
* Work you're starting at the end of the day to read tomorrow
* Anything where the check-ins are pure overhead
* A task that needs several rounds of "now do the next one"
## When not to
* The finish line is vague — it'll keep going longer than you meant
* Each step needs a judgement you'd want to make
* It's the first time this coworker has done this kind of work
* The work is irreversible and you haven't [planned](/cowork/chat/plan-and-build) it
**Give it a finish line.** "Keep going until every ticket has a draft reply" is a target. "Work on the tickets" isn't, and it's the difference between a run that ends and one you have to stop.
## What it doesn't change
Running long doesn't mean running unsupervised.
* **[Approvals](/cowork/control/approvals) still pause it.** It waits for your yes exactly as it would otherwise.
* **[Permission rules](/cowork/control/permission-rules) still apply.** Nothing is loosened because it's in a long run.
* **You can still interrupt.** Type a correction and it's picked up at the next step.
So a long run is your coworker not stopping to *ask whether to continue*. It isn't your coworker acting with fewer controls.
## Good to know
This is one of the [built-in skills](/cowork/skills/built-in) — every coworker knows how to do it, with nothing to switch on.
A long run costs more than a short one, simply because more work happens. Worth knowing before starting one on a big list.
Pair it with [Thorough review](/cowork/chat/thorough-review) for a long job you won't watch. A mistake early in an unattended run otherwise carries through everything after it.
## Keep working, schedules, or self-scheduling?
| | What it is |
| ----------------------------------------------------------- | --------------------------------------- |
| **Keep working** | Push through one job now, in one run |
| **[Schedules](/cowork/autonomy/schedules)** | Start fresh work on a timetable |
| **[Self-scheduling](/cowork/capabilities/self-scheduling)** | Pause, and come back to this task later |
## Common questions
Type `/stop`, or stop the task. Everything finished so far stays.
Yes. It keeps working and the whole run is there when you come back.
It tells you rather than looping on the same failure. A run that can't proceed ends and explains why.
No. It stops when the job is done — which is why writing a clear finish line matters.
Yes. A scheduled task can itself run long, which suits a nightly job that has to work through a queue.
## Next steps
How to start and stop it.
Following a long run.
When one coworker isn't fast enough.
Everything that runs without you.