Cron for the rest of us: scheduled tasks by text

The original "computers can do this for you" feature is the scheduled job. Cron has been around since 1975. Calendars have had recurring events since the 1980s. Phone alarms. Slack reminders. Apple's Calendar app. The unifying property of every prior version of this is that the user has to know a specific syntax (cron expressions, recurring-event rules, alarm presets) and click through a UI to set it up. The friction is not where the schedule lives; it is in the telling.

A scheduled task in a personal AI agent is the version of this where the friction is gone. The user describes the schedule in plain English. The agent translates to a cron expression internally, sets the task, and confirms. Done. No syntax, no panel, no "advanced settings." This post is a tour of the surface and the recipes most of our users end up keeping around.

The four shapes of schedule

In a strong agent, every recurring or future task fits one of four shapes:

The user almost never uses ISO timestamps. They use the first three, in plain English. The agent's job is to figure out which shape and what arguments. This is one of the genuinely-easy problems for a frontier model: scheduling syntax is well-defined, and natural-language-to-cron is essentially solved.

The recipes most users end up with

A representative power user's scheduled-task list, after a month with the agent:

Every weekday at 7:00 AM ET → morning brief on WhatsApp
Every Sunday at 8:00 PM ET → next week brief
Every Friday at 5:00 PM → weekly retrospective summary
On the 1st of every month at 9:00 AM → credit score check
On the 28th of every month at 9:00 AM → next-month bills preview
Every 4 hours during business hours → check inbox for urgent client emails
In 30 minutes → remind me to take out the trash
On 2026-12-15 at 9:00 AM → annual KYC review

Eight tasks, each one a single sentence to set up. The morning brief is the keystone: most users start with that, then add the next-week brief on Sundays, then a finance check at the start of the month. The "in 30 minutes" reminders are constant; people use them like timers, dozens of times a week.

How it actually fires

When a scheduled task is due, the agent runs it as a full agent invocation. That is important: it is not a static notification. The agent has access to its tools, its memory, and its skills. So "send me a morning brief" pulls fresh data from your calendar, inbox, and tasks every time, applies your tuning ("skip newsletters"), and sends the result. The schedule fires the same way you would fire a chat turn, just without your typing it.

This means a scheduled task is "smarter" than a notification. It can decide not to send if there is nothing meaningful to report. It can choose the channel. It can include a payment link, a calendar invite, or an attachment. The format is whatever the agent decides is right at the moment of firing.

Pause, resume, and one-off override

Three control verbs, all in plain English:

The pause / resume / trigger surface is the second-most-used after creation. The third-most-used is patching ("update the morning brief to skip newsletters"). The agent versions every patch automatically, so reverting is one sentence away.

Where it gets interesting: skill-bound schedules

The plain-English layer is the easy part. The interesting part is when a scheduled task is bound to a skill that knows your specific preferences. Two examples:

The Sunday brief. A weekly summary. The default version is "what did you do, what is open, what is next week." The user-tuned version is "what did you do, what got skipped, what is next week, who do you owe a reply to, what bills hit on Tuesday, and any health-data outlier worth flagging." The Sunday brief skill carries that tuning, so each Sunday at 8 p.m. the agent does the right shape of work.

The morning brief during a launch. During the week of a product launch, the morning brief gets a launch-specific section: yesterday's signups, support tickets, churn signals, social mentions. The user did not have to add a new task; they just told the agent "for this week, include launch metrics in the morning brief." The skill takes the patch, the schedule keeps firing, the brief includes the new section automatically.

Failure modes

Two:

Time zones. The agent stores tasks in UTC and fires them in your local time. The first time you travel and a task fires at the wrong hour, you tell the agent your new time zone. From then on it adjusts automatically.

Silent failures. A scheduled task that fires when the agent has no internet connection loses the firing window. The system catches this on the next check, but a small lag is possible. For time-critical tasks (a meeting recap that needs to land before the next meeting), use one-shot tasks with a small buffer rather than tight intervals.

The first scheduled task to set up

Set up the morning brief on day one. Even if you change everything else later, the morning brief is the keystone routine that converts the agent from "a tool I sometimes text" to "the assistant who runs my mornings." Text "send me a morning brief every weekday at 7am" to +1 (470) 821 4265.