Scheduled tasks: work that is already done before you start
Most automation that pays off is not complicated. It is one instruction you set up once that runs every week without you. How it works, and where it goes wrong.
New to this topic? Start with How to Build Your First Agentic Workflow with dGENIX
Automation rarely starts big
When people think about automation they picture a system: something that has to be built, with steps and conditions and a diagram next to it. That exists too, and sometimes it is what you need.
But most of the time you get back comes from something far smaller. One instruction you now carry out yourself every Monday morning, that you set up once and never touch again.
That is what a scheduled task is. You write down what should happen, you choose when, and your assistant carries it out exactly as if you had asked at that moment.
What you can put in one
You write the instruction the way you would message a colleague. There is no language to learn and no screen of blocks to drag around.
A few that hold up in practice:
- Every Monday at 08:30, an overview of your calendar for the week ahead
- Every morning at 09:00, three priorities based on what came into your inbox
- Every Friday at 17:00, a weekly report from your meetings and completed tasks
- On the first of the month, a financial overview from a Google Sheet
- Every Monday, fresh social visuals made in your own studio
Five frequencies are available: once, daily, weekdays, weekly and monthly. Weekdays is there for a reason; a daily briefing that lands in your inbox on Sunday morning is one you learn to ignore within two weeks.
The difference between a good and a bad instruction
This is where all the craft is, and it comes down to precision.
| Weaker | Better |
|---|---|
| "Make a report" | "Make a weekly report of my meetings from last week, with a section per day" |
| "Check my mail" | "Find what in today's mail needs a reply from me, and say why for each one" |
| "Keep an eye on the site" | "Run an SEO audit on my homepage and name the three heaviest findings" |
The right column is not better because it is longer. It is more specific about the outcome. Your assistant runs the instruction without you there to steer, so anything you leave open gets filled in by him. That is amusing exactly once.
A practical test: read your instruction back and ask whether a new colleague could carry it out without asking anything. If not, something is missing.
What it can and cannot reach
A scheduled task has the same access you have in a normal conversation. No more.
Yes: your connected services, the skills you switched on, and your knowledge document with your settings.
No: anything you did not give access to. If a connector is missing, it says so rather than guessing.
That sounds obvious but has a consequence that surprises people: a task that worked last week can go quiet this week because you disconnected something. The task is not broken, it just cannot reach somewhere any more.
The trap everyone walks into
The most common mistake is not technical. It is setting up too many.
Start enthusiastically and within a week you have seven scheduled tasks all dropping something in your inbox. After two weeks you stop reading them, and then the effect is the exact opposite of the intention: you increased the noise instead of reducing the work.
A better approach is boring and it works:
- Start with one task, for something you genuinely do yourself every week
- Let it run for two weeks and check whether you use the output
- If you do not, adjust the instruction or switch it off
- Only when it works, add the second
You can pause, edit or delete a task without losing anything. Pausing is almost always the better first move when something disappoints: then you can still read the instruction back.
What a scheduled task is not
It is not a workflow. An agentic workflow is a series of steps that hang together with decision points in between. A scheduled task is one instruction at a fixed time.
That distinction is not academic. If you notice yourself writing an instruction containing "and then" and "unless", you are building a workflow in a field that was not made for it. That is when the other side of the platform is the right place.
The reverse happens more often: people build a workflow for something that is one instruction. That takes longer to set up and breaks more often.
Why this pays off more than it looks
The gain is not in the minutes the task itself takes. Making a calendar overview takes three minutes, so fifty weeks is two and a half hours. Nice, nothing more.
The real gain is in what stops going wrong. Work that happens at a fixed moment does not get skipped in a busy week. The report that is always there gets read. The monthly check that is never forgotten finds things while they are still small.
That is the same reason a recurring SEO check is worth more than one thorough audit you do once. Not because the audit is worse, but because a site changes and your attention is not on the same spot every month.
Getting started
Go to Dashboard, Tasks and create one. Give it a name you will still understand in three months, write the instruction as a message, pick a frequency and switch it on.
Read the first result. Not to check that the system works, but to see whether you wrote the instruction sharply enough. That is almost always where the problem is, and it takes thirty seconds to fix.
