Putting Agents on a Clock

The best work an agent does is the work you never asked for, because it was already on the calendar. Scheduling is how your system stops waiting for you and starts running on its own, so you wake up to work that is already done.

9.10.1Agents that run without youAdd note

Until now, every agent waited for you to press go. Scheduling flips that. You put an agent on a clock and it runs on its own, on time, whether or not you show up.

That is the moment the system stops being a tool you operate and starts being one that operates, doing the rounds because the clock struck, not because you remembered.

9.10.2Put the routine work on a clockAdd note

Look at everything you do on a rhythm: the weekly numbers, the daily health check, the monthly money review. Every one is a scheduled agent waiting to exist.

You set its cadence with a cron line, five fields that mean "run at this time":

Write the job once, give it a cadence, and it runs forever. The work that is important but never urgent, and so never gets done, finally does, because a clock does not procrastinate.

That format is cron, which is built into Mac and Linux. Windows has the same thing under a different name, Task Scheduler. You do not set either up by hand: you tell your agent the cadence and let it wire the right one for your machine.

9.10.3Schedule it, or trigger it on demandAdd note

The same job has two doors. On a schedule it runs every morning at eight; on demand you run one command and it goes now. You want both: the schedule for the rhythm, the manual run for the moment you need the report early.

A cron line is only the clock. What it fires is one command that runs your agent with no chat window, reading the job's instructions from a file. Your agent wires this, but you should recognize it:

The schedule fires that exact command for you at the set time. A plain cron runs it on your machine; a bigger engine handles it when you have many, which is the next choice to make.

9.10.4Pick the tool that fits how you runAdd note

Plain cron is the floor: it fires jobs while your machine is on, with no dashboard. One step up is an orchestrator, a tool that runs your jobs unattended and gives you a dashboard to see every run, retry, and failure at a glance. Prefect, Airflow, and Temporal are the common ones, and Prefect is the gentlest to start with. Some agent tools also ship their own scheduler, so check what yours already offers before you add anything.

Which one fits depends on four things: how often the job runs, whether it must run without you opening any app, whether you want a dashboard, and where it lives. This space keeps moving, so pick what fits today and swap later.

9.10.5Make an unattended run safeAdd note

A job that runs while you sleep needs more than a clock, or it fails in the dark and you find out late. Three things make an unattended run trustworthy:

1. Gate: before it acts, the agent confirms it is still allowed to (the guardrails of the next chapter). It fixes what is safe and escalates the rest, never spending or deleting on its own.
2. Record: every run writes what it did to one log (you build this in the ledger chapter), so "what ran last night" has one honest answer.
3. Watchdog: something tells you when a run stops firing. A scheduler does not fail loudly; it goes quiet, so you watch for the missing run, not just for errors.
4. Cap: bound what one run may spend and how many times it retries. An agent left to loop is an open wallet, so a cheap no-AI watcher that wakes the costly agent only when there is real work keeps the bill small.

9.10.6Know the bill before it arrivesAdd note

A fleet on a clock spends money every time it wakes, whether or not there was anything to do. That bill is the most common way a first scheduled system ends, and it is entirely preventable.

Match the model to the job. A reporter that summarizes a file runs fine on the cheapest small model; an operator that changes things does not. That one choice moves the bill more than anything else you can do.

Then set the cap where the agent cannot argue with it: a hard spend limit and a usage alert in your model provider's own dashboard, not in a prompt. Read the actual bill after the first week, before you put a second job on the clock.

9.10.7Wake up to finished workAdd note

This is the payoff, and it changes how the whole thing feels. You open your laptop and the reports are already written, the checks already run, the problems already flagged or fixed.

You are not starting the day's work. You are reviewing work that already happened. The system ran a full shift while you slept.

9.10.8Put one job on a clockAdd note

This prompt turns a routine chore into a scheduled agent:

Do this now: paste the prompt, pick one thing you do on a rhythm, and let your agent turn it into a job on a clock.