On this page
The schedule never starts a mission, never approves the plan or the completion, and never sends any of the messages your team drafts. Those three assurances hold without exception and cannot be configured away. Two things do run alongside them: automatic steps from your template and — if you switched it on — automatic field output. Both are covered in the section “What the schedule does”.
This function is new and has not yet run continuously in practice. Switch it on for a single, uncritical mission first and look at the log under “Runs & Cost” the next day before you use it more widely.
Where the switch is
The “Keep working on its own” switch sits on the mission detail — not in the mission list. So you have to open the mission. Seeing and changing it requires the “Manage missions” permission. It is off by default; you switch it on per mission, not for the whole organization, so the decision stays attached to each individual piece of work. It is still there on a finished mission — and it matters there, because drawing the lessons is exactly what the schedule still does at that point. Only a cancelled mission has no switch, because the schedule never picks one up again.
On a mission still in draft the switch has no effect yet — the note beside it says so. Start the mission first; only a running mission can be continued.
What the schedule does
The schedule runs in the background and checks every few minutes whether a mission has work. It picks up the same mission at most every ten minutes — so it does not rush it. When it finds work, it does exactly what you would otherwise trigger by clicking.
- Setting up the case
- You approved the plan, but the case was never created — the classic “closed the tab right after approving” case. Your yes is on record, only the motor was missing. The schedule creates the case and delegates the tasks.
- Executor work
- The case is running and there are open tasks an agent can work itself. The schedule does the same as your click on “Let the team work” — completing tasks, writing field values, drafting messages.
- Reflection
- The mission is completed but the lessons were never drawn. The schedule catches up on that so the experience isn't lost.
Automatic steps that are part of your template — a “send email” action, say — continue perfectly normally. The schedule creates the case and the rest works as always. The difference concerns only what an agent proposes itself.
Which leads to two cases that are easy to miss. First: if the schedule sets the case up and your template contains a send step with the trigger “Send automatically”, that message goes out with no further involvement. Second: if this mission's field output is set to “Automatic”, the run writes the proposed values straight into the case without you having seen them. Neither was decided by the schedule — you set both beforehand. Go through both before you let a mission run unattended.
What the schedule never does
- It starts no mission. The first run is always your click on “Start mission”.
- It approves neither the plan nor the completion, and no send proposal. It puts field suggestions to you as well — unless you set this mission's field output to “Automatic”. Then the run writes the values with no intermediate step, with nobody having the app open.
- It sends no message an agent proposed. Not even one you have already approved goes out through it. Automatic send steps from your template are a separate matter — those do run on.
- It does not touch a mission at all while a plan or completion approval is sitting with you. If you have the ball, it stays out of it entirely.
The most important pitfall: once you approve a send proposal, the schedule will NOT send it. The message waits until you open the mission and let the team work once. The mission tells you so — but if you close the tab after approving and trust the schedule to handle it, the message waits.
On top of that, two latches on the lead agent stop the schedule immediately: if the lead agent is deactivated, all its missions are frozen. And if it lacks the “Run mission” tool, nothing runs at all — on the click-free path that is the only switch that bites per agent. Both are an effective kill switch if you want to stop an autonomous mission quickly.
The latch: when nothing moves any more
A mission that isn't going anywhere should not run forever burning budget. So bricksta counts: if three runs in a row fail to advance the mission, it pauses itself. It is then marked “Paused” in the mission list, and the mission detail says why.
What is counted is the attempt, not the result. A run that crashes before it can report anything counts too. That is deliberate: otherwise a broken mission would run the same expensive race every ten minutes forever, without leaving a trace anywhere.
The counter starts over as soon as a person has decided something since the last attempt. That is the everyday case: with the default “Review” setting for field output, nearly every run eventually waits on you — without this rule the mission would constantly run into the latch.
Why a mission pauses — and what helps in each case
The mission detail states the reason plainly. These situations exist:
- There was nothing to do
- The schedule picked the mission up three times and found no task an agent could work itself. Way out: check in the case who the open tasks are assigned to. Usually they hang on people — or on an agent you haven't yet given the “Work mission tasks” tool.
- Something is waiting for your review
- There are field suggestions or another decision on the mission detail. Way out: decide them — approve or discard. After that the mission picks up again by itself; you don't need to touch the switch.
- An approved message is waiting to be delivered
- You approved a send proposal, and the schedule never sends. Way out: open the mission and press “Let the team work” once — then the message goes out.
- The runs were denied
- Usually the budget: the gateway turned the run away before anything could happen. It can also be a deactivated agent or a missing tool. Way out: fix the cause as described in the article Budget, cost and runs, then switch the toggle off and on again.
- The runs failed
- A substantive error, at the AI provider for instance. Way out: open the mission and press “Let the team work” to see which error message actually comes back — in the foreground you get more to see than in the log.
- The runs aborted before they could report anything
- Almost always a run that takes too long — typical for a plan with many tasks. Way out: open the mission and let the team work once by hand, or break the plan into smaller steps next time.
If there is nothing to decide and you want to try again anyway: switch “Keep working on its own” off and on again. That resets the latch.
The log: what happened while you were away
What ran without you is in the Agent Center under “Runs & Cost”, in the “Ran on its own” list. It requires the “Manage organization” permission and shows the latest 50 runs; if older ones are hidden it says so below. Clicking the mission takes you straight there.
Per run it shows the mission, the time, the phase (setting up the case, executor work or reflection) and how it went:
- Moved forward — something was really done. Beside it stands what: completed tasks, fields written, messages sent, tasks created, lessons drawn.
- Nothing to do — the run found no work. Once is unremarkable; three times in a row pauses the mission.
- Waiting for you — something is up for review on the mission detail. Beside it stands how many.
- Waiting to send — an approved message is ready and the schedule does not send.
- Denied — the gateway turned it away, with the reason spelled out beside it.
- Failed — the run failed substantively.
If the list is empty, no mission has worked on its own yet — missions only run by themselves once you switch that on for them. An empty list is therefore not a fault but the normal state before you flip the switch for the first time.
When it is worth it — and when it isn't
The schedule pays off where a mission has many small steps that could each move on immediately: you approve a plan in the morning, and by midday the case is set up and half worked through, without you having to click in between.
It is not worth it on a mission whose tasks mostly hang on people anyway, and it is not worth it on something you want to accompany step by step regardless — there it merely produces idle runs until the latch trips. And it does not replace supervision: it does nothing you haven't approved, but it also won't tell you anything of its own accord. Checking the log stays your job.
A good way in: switch it on for a mission you are running anyway, and the next day compare the log with what you would have expected. After that you know whether your missions are of the shape that benefits.