# How to Run AI on a Schedule Instead of Prompting It

*How-to — 2026-09-15 — by Mahmoud Zalt*

Set a time and a standing brief once, and your AI Employee runs the same job on repeat with no prompt from you, leaving a report to read later.

**Short answer.** **You run AI on a schedule by attaching a time and a frequency to a job instead of typing it again every week.** Write the instructions once, pick the cadence, set the line between what it may do alone and what it must ask about, and the job repeats itself. Sistava calls this a recurring task, and every run leaves a written record. Plans start at 25 per month with credits included.

You already know which jobs repeat. The Monday pipeline review. The weekly competitor check. The month end summary you rebuild from scratch every time. You have probably even written a great prompt for one of them, saved it in a note, and pasted it back in week after week.

That paste is the part worth deleting. Not because typing is hard, but because it makes you the trigger. If you are busy, sick, travelling, or simply forget on a Monday, the work does not happen. A saved prompt still needs a human to start it, so it is not really automated. It is just organised.

A schedule removes you from the middle. In **Sistava**, you hire an AI Employee and give it a recurring task: the instructions, the cadence, the tools it may use, and the limits it must respect. From then on the clock starts the work. The Employee already carries a standing brief about your business, so it does not need context pasted in, and it writes each run into a work journal so you can read what happened without having watched it.

## At a Glance

- **Once** How many times you write the instructions
- **Hourly to monthly** Cadences a recurring task can use
- **Every run** How often the work journal is written
- **25/mo** Entry plan with credits included

## What does running AI on a schedule actually mean?

It means the job has a time attached to it. You define the work once as a standing instruction, choose how often it should happen, and the AI Employee wakes on that cadence, does the work, and stops. You are not present, and you do not need to be, because everything it needed was decided before the first run.

The mental shift is small but total. A prompt is a request, and a request needs a requester. A schedule is a standing job, and a standing job needs an owner. You stop being the person who asks for the work and become the person who reviews it, which is the same move you make when you hire someone and stop doing the task yourself.

## Benefits

### The instruction

What to do, written the way you would explain it to a new hire on their first Monday. Specific enough that two people would do it the same way.

### The cadence

How often and at what time. Every weekday at 6am, every Monday at 7am, the first of the month. This replaces you remembering.

### The boundary

Which actions it may take alone and which queue for your approval. Sending, spending, and anything a customer will see usually start on the approval side.

### The record

Each run writes what it did, what it found, what it decided, and what it skipped. That entry is what you read instead of watching.

## Why does a schedule beat a really good prompt?

Because a good prompt still dies the moment you get busy. The quality of the writing does not matter if nobody pastes it in. A schedule survives your bad weeks, your holidays, and the Monday where three things caught fire before 9am, and that reliability is worth more than any clever phrasing.

There is a second difference that people miss until they feel it. A pasted prompt starts from zero every time, so it never learns your business. A scheduled job on an AI Employee carries the standing brief and the history of previous runs, so week four knows what week three found. The work compounds instead of resetting.

And a schedule creates a rhythm you can plan around. When you know the pipeline note lands at 7am on Monday, you stop building your morning around producing it and start building it around acting on it. That is the whole point. The output stays the same, the hour it costs you does not.

## What kind of work belongs on a recurring schedule?

Work that repeats with the same shape and different contents. Gathering, checking, sorting, drafting, and summarising all fit, because the method stays constant even when the material changes. If you can describe the job as a procedure rather than a decision, it belongs on a clock.

Judgment work does not belong there. Whether to raise your prices, whether to fire a client, whether to take on a project that feels wrong, none of that gets better by happening at 6am without you. Put the preparation on a schedule and keep the decision for yourself, and you get the speed without giving up the steering wheel.

| Put it on a schedule | Keep it in your hands | Why |
|---|---|---|
| Weekly pipeline and follow up sweep | Deciding which deal to walk away from | The sweep is a procedure, the walk away is a judgment call |
| Daily inbox sorting into needs you and handled | Replying to the client who is upset | Sorting has rules, an upset client needs your voice |
| Monthly competitor and pricing check | Changing your own prices | Gathering the facts repeats, acting on them does not |
| Drafting recurring reports and status notes | Signing anything or committing money | Drafting is reversible, commitments are not |

## A worked example: the recruiting agency that starts Monday already done

Hannah runs a four person technical recruiting agency in Manchester. Her Monday used to disappear into the same three hours of preparation: pull every open role, check which candidates had gone quiet, and rebuild the client update her team reads at the 10am stand up. She never enjoyed it and she was never late with it, which is exactly the kind of task that quietly eats a career.

She wrote a standing brief first: her nine open roles, her four clients, how she describes candidate stages, and one hard rule that no candidate or client is ever contacted without her approval. Then she made one recurring task set to every Monday at 6am. Pull every open role and its stage. Flag any candidate with no contact for eight days or more. Draft a nudge for each one. Write the whole thing into a single client update.

The first two Mondays she left it on report only, so it drafted and sent nothing. Both times she found small things to correct, mostly wording and one stage she had described sloppily. By week three the update was landing the way she would have written it, so she let the nudges go out to candidates while keeping client messages on approval. Her Monday preparation went from three hours to about twenty minutes of reading and clicking.

The part she did not expect was the compounding. Because each run could see the previous one, week six flagged that one client's roles had been stalling at the same stage for a month. Nobody asked it to notice a pattern. It could only notice because the job kept running and kept a record, which a pasted prompt would never have had.

## What a schedule will not fix

It will not fix a vague instruction. A schedule is a multiplier, so a fuzzy job description gets you fuzzy output every single week instead of just once. If you cannot explain the task clearly enough for a competent new hire to follow it, do not put it on a clock yet. Write it properly first.

It also will not invent facts it was never given. Anything that lives only in your head, the client who hates phone calls, the supplier you stopped trusting, the rule you never wrote down, is invisible until you put it in the brief. And it will not make judgment calls for you. A scheduled job that reaches something outside its rules stops and records why rather than guessing its way forward, which is the behaviour you want when nobody is watching.

**Start with one job, not five.** Pick the single task you most resent repeating, put it on a cadence, and leave it report only until three clean runs in a row. Adding a second job before the first is trustworthy is how people end up with five broken schedules and no idea which one to fix.

### Setting up your first recurring task

1. **Write the standing brief** — Explain your business once: what you sell, who your customers are, how you sound, and what is never allowed. Every run carries it, so you never paste context again.
2. **Describe the job like a procedure** — Steps, not wishes. Pull this, check that, flag anything matching this rule, write it up in this shape. Specific enough that two people would produce the same result.
3. **Choose the cadence and the time** — Match it to when you need the output, not when you are awake. If you read it at 8am on Monday, run it at 6am on Monday.
4. **Set it to report only** — For the first runs, tell it to draft and record rather than send and act. You see exactly what it would have done with none of the consequences.
5. **Cap what a run may spend** — Give the task a credit budget so an unusual week cannot become an unusual bill. When the cap is hit the run stops and tells you why.
6. **Read three runs, then widen** — Correct what is off in the brief rather than in a one off message. After three clean runs, move the safest actions from report to act and leave the rest on approval.

A clock is only one way to start work. If your jobs should begin when something happens rather than at a fixed hour, read [how to get AI to work without being asked](/insights/how-to-get-ai-to-work-without-being-asked). For the overnight version of the same setup, see [how to make AI work while you sleep](/insights/how-to-make-ai-work-while-you-sleep).

## Frequently asked questions

## FAQ

### How often can an AI Employee run a scheduled job?

From hourly through daily, weekly, and monthly. Most people find that daily is right for sorting and drafting work, weekly is right for reviews and summaries, and hourly is only worth it when something genuinely changes that fast, like an inbox during a launch. Every cadence costs credits per run, so a faster schedule is a real choice rather than a free one.

### Do I need to know how to code to schedule AI work?

No. You describe the job in plain English and pick a time from a list. There is no cron syntax, no workflow builder with boxes and arrows, and no script to maintain. If you can write an instruction for a new colleague, you can write a recurring task.

### What happens if a scheduled run fails or hits something unexpected?

It stops and records the reason rather than guessing. A tool that is offline, a rule it is not allowed to break, or a situation the brief never covered all end up in the run record as an unfinished item with the cause attached. The next scheduled run still happens on time, so one bad night does not break the rhythm.

### Can I change the instructions without rebuilding the whole task?

Yes, and you should. Editing the standing brief or the task description is the normal way to correct behaviour, and the change applies from the next run onward. Fixing the instruction is always better than sending a one off correction, because the one off is forgotten by the following week and the instruction is not.

### Can two scheduled jobs run at the same time without clashing?

Yes. Each task keeps its own instructions, its own cadence, and its own budget, so you can pause or edit one without touching the others. Keeping jobs separate is also easier to review, because a run record that covers one job tells you clearly whether that job worked.

Running AI on a schedule is less a technique than a decision about who starts the work. Write the job down properly, give it a time, keep it report only until it has earned more, and let the record be what you read instead of the process being what you run. The saved prompt in your notes app was never the problem. Being the one who had to paste it was.

**Tags:** ai-on-a-schedule, recurring-ai-tasks, scheduled-ai-employee, automation-without-prompts, hire-ai-employee