AI agent scheduled tasks are instructions your agent runs by itself at a time you choose, such as a morning plan or a Friday recap, with the result delivered to the chat you already use. The ones that get read are narrow: one message, one purpose, a fixed time.
A scheduled check-in is the simplest kind, a message your agent sends without being asked, so it reaches you instead of waiting to be opened. The version that fails is a daily digest of everything, which nobody reads after the third day. Below are three check-ins worth setting up and how to word the instruction so the output earns its place.
How does an AI agent scheduled task work?
You give the agent an instruction and a time. It runs the instruction on schedule and sends you the result: every morning, every Friday, or once at a time you specify.
The result usually lands in your existing chat, such as Telegram, which is why this works better than another app: the message arrives where you already look. With Able°, your agent's scheduled messages arrive in your Telegram or LINE chat. Chat tools have the same idea; ChatGPT, for example, calls them scheduled tasks.
One detail matters more than people expect. In many agent engines, each scheduled run starts as a fresh session without your recent conversation. The Hermes Agent docs say the instruction "must contain everything the agent needs". So a scheduled task works from its instruction and from what the agent has saved to memory, and it won't know what you happened to mention in chat that afternoon.
You don't need to build anything: a clear instruction and a time are enough.
A morning plan
Send it shortly before your day starts. It should cover the fixed commitments you've told the agent about, the one thing that matters most if the agent can tell, and anything still waiting on a decision.
"Every weekday at 7:30 London time, send me a short message with today's commitments, one thing to prioritise, and anything you've noted as still unresolved. Keep it under 120 words. No greeting."
Those last two sentences do a lot. Without a length limit, the message grows week by week. Without "no greeting", you get a paragraph of warm-up you'll learn to skip. For a fuller approach to the day, see Planning your day with an AI agent.
An end-of-day prompt
This one arrives as your working day ends and asks you questions instead of reporting: what moved today, what didn't and why, and what comes first tomorrow.
"Every weekday at 18:00, ask me three short questions: what moved today, what didn't, and what's the first thing tomorrow. One line each. Don't summarise my whole day."
It takes thirty seconds to answer. If you ask the agent to note your answers, you build up a record you can draw on for Friday's recap.
A Friday recap
Set it for Friday afternoon, or whenever your week ends. Ask for decisions made this week, items still open with how long they've been open, and one pattern worth noticing.
"Every Friday at 16:00, send me a recap: decisions from this week, open items with the number of days they've been open, and one thing that seems to be slipping. Under 200 words."
The line about something slipping is where most of the value sits. It's the observation you'd make if you sat down and thought about the week, and Friday afternoon is when you're least likely to do that.
How do you write a good scheduled instruction?
State the time, the content, the length and the style, in that order.
1. The time and frequency. Name the days and include your time zone. "Every weekday at 7:30 London time" is clearer than "regularly". Agents often run on a server somewhere else, and some engines use the server's own clock unless a time zone is set.
2. The content. Name the two or three actual items you want, rather than a topic.
3. The length. Give a word limit or a line count. Nothing else you add will do as much.
4. The style. "No greeting", "bullets only", "no summary of things I already know".
A quick test: if the message arrived and you had thirty seconds, would you get what you needed? If not, cut something.
Why do scheduled messages get ignored?
Usually because they say too much, ask for nothing, or arrive at the wrong time.
Too much. A daily report covering everything becomes noise. Three items beat fifteen.
Nothing to act on. A message you can't act on is just more reading. End every check-in with something to do, decide or answer.
Wrong time. A morning plan arriving at 11 is already out of date. Schedule it for when you really look at your phone.
Nothing changes. If the same content arrives every day, all it tells you is that nothing is happening. Change the prompt or stop it.
Never adjusted. The first version is rarely right. Expect to reword it two or three times in the first fortnight. Setting one up in your first week is a good way to start that cycle.
A small checklist
- One purpose per scheduled message
- Specific days, time and time zone
- Two or three content items, named
- A word or line limit
- A style instruction, including what to leave out
- Ends with something to decide or answer
- Reviewed after one week and adjusted
When a schedule won't help
If your days are unpredictable, a morning plan will often be wrong by 9am. A single evening prompt suits that kind of week better than a schedule that assumes structure.
If you already keep a notebook or task list you trust, a scheduled message may only duplicate it. Keep a check-in only if you act on it. And if you know you won't read it on a busy day, don't set it, because skipped messages soon turn into a habit of skimming past everything the agent sends.
Remember, too, that a scheduled message only knows what your agent knows. A meeting you never mentioned won't appear in the plan, and a plan that ranks the wrong thing first is a suggestion you're free to overrule.
Common questions
Can an AI agent send you reminders?
Yes, if it supports scheduled tasks. You can usually ask in plain language for a one-off reminder, such as "remind me at 3pm to call the supplier", or a recurring one, such as every Monday morning. The reminder arrives as a message in your chat app. Include the time zone and enough detail that the reminder still makes sense on its own.
Does a scheduled task remember what you talked about earlier?
Often not directly. In many agent engines, each scheduled run starts fresh, without the conversation you had earlier that day. It relies on the instruction itself and on anything the agent has saved to memory. If you want a check-in to pick something up, ask the agent to note it, or put the detail into the instruction.
Can you change or stop a scheduled task?
Yes. You can normally ask the agent to change the time, reword the instruction, pause it or delete it, in the same chat where you set it up. Ask the agent to list your active scheduled tasks every so often, so old ones don't keep running after they've stopped being useful.
Do scheduled tasks use AI credits?
Yes. Each scheduled run is a normal request to the AI model, so it uses tokens, credits or plan usage in the same way as a message you send yourself. A short daily check-in uses little. Long instructions, long outputs and very frequent schedules use more, so keep check-ins short and remove the ones you no longer read.