Emerging

Scheduled agent runs

Recurring autonomous work on a visible schedule, with results delivered into an existing surface.

MScheduled runsThu 1 Oct · 4 schedules

Try it: Run the Monday brief now and see where it lands. Then open the report Milky paused because nobody was reading it.

In the wild

How 3 products ship this pattern. Click a screenshot to enlarge it.

ChatGPT Scheduled page with a Schedule a task input and example tasks: Daily brief, Email monitor, Industry trends. — ChatGPT Scheduled
A dedicated home for recurring runs, seeded with examples that show the cadence in plain language ("Every Monday…").
Source: OpenAI Help Center ↗

ChatGPT recurring task detail showing the prompt, a Repeat frequency set to Custom, an Until setting of Never, and a pause button. — ChatGPT task detail
Each task exposes its prompt, frequency and end condition as editable fields, with pause always one tap away.
Source: OpenAI Help Center ↗

Manus Scheduled Tasks calendar for April 2026 with recurring agent tasks such as 'Weekly Report Digest', 'PR Review' and 'Competitor Analysis' at set times — Manus
Scheduled tasks appear on a calendar with their run times, so recurring agent work is visible in advance and can be edited or paused.
Source: Manus blog ↗

When to use

Monitoring, reporting and maintenance jobs where the value is in never being asked for.

Seen in

ChatGPT scheduled tasks, Zapier scheduled agents, Deep Research recurring briefs.

For engineers

Show last-run and next-run; auto-pause on repeated unopened results or repeated failure.

Build prompt

A neutral, component-only prompt for a coding agent. Edit the stack block or anything else to fit your project, then copy it.

# Build the "Scheduled agent runs" pattern: `ScheduledRunList`

## Stack (edit this block to match your project)
- React 18+ with TypeScript: one function component, typed props, no `any`.
- Tailwind CSS, using your design system's tokens if you have them.
- Presentational only: data comes in through props, decisions go out through callbacks. No fetching or global state inside.
- If your codebase already has conventions (naming, shared components, tokens), follow them over this spec.

## What the pattern is for
Recurring autonomous work on a visible schedule, with results delivered into an existing surface.

Use it when: Monitoring, reporting and maintenance jobs where the value is in never being asked for.

## What to build
A list of recurring agent jobs, each showing its plain-language cadence, last run, next run, where results are delivered and whether anyone is reading them. A person can run, pause, resume, edit or end a schedule, and sees when the system auto-paused one. It renders as a settings page or a panel next to the surface that receives results.

### Props
- `schedules`: Array<{ id: string; title: string; instruction: string; cadence: string; deliverTo: string; status: "active" | "paused" | "auto-paused" | "ended" | "running"; lastRun?: { at: Date; outcome: "delivered" | "failed"; opened?: boolean; openedBy?: number }; nextRunAt?: Date; endsAt?: Date; endCondition?: string; autoPauseReason?: "unopened" | "failures"; costPerRun?: string }>.
- `selectedId?`: string and `onSelect(id | null)`.
- `onRunNow(id)`, `onPause(id)`, `onResume(id)`, `onEnd(id)` and `onEdit(id)`: callbacks. The component never runs or schedules work itself.
- `autoPausePolicy?`: string, shown in each row, e.g. "Pauses after 2 unopened results".
- `emptyAction?`: { label: string; onClick: () => void }, for creating the first schedule.

### States it must render
- Active: title, cadence ("Every Monday at 08:00"), destination, "Last run Mon 28 Sep, opened by 4" and "Next Mon 5 Oct, 08:00".
- Running: "Running now" with the destination it will land in; the next-run time stays visible.
- Auto-paused: amber row stating why ("Paused: the last 2 results were not opened" or "Paused after 3 failed runs"), with Resume and End; no next run is shown.
- Paused / ended: muted row with who paused it or the end date; ended schedules move to an "Ended" section.
- Empty: a short explanation of what schedules do with the create action.

### Interactions
- Run now calls `onRunNow` and the row shows Running; on delivery it shows "Delivered just now" with a link to the destination.
- Pause is one click from the row, never only inside an edit form.
- Resume on an auto-paused schedule asks for an end date or condition first (optional), so it does not run indefinitely again.
- Selecting a row expands the full instruction, end condition and cost per run.

### Edge cases
- A schedule with no end date or condition: show "No end date" as a visible attribute, not blank space.
- Last result delivered but not opened: show "Not opened" in the last-run line before auto-pause triggers.
- The next run falls inside a pause window or after `endsAt`: show "No more runs" instead of a date.
- Time zones: show times in the viewer's zone and include the zone when the schedule's own zone differs.

## Must avoid
Schedules that keep firing after the reason expired, producing output nobody reads and cost nobody sees.

## Engineering constraint
Show last-run and next-run; auto-pause on repeated unopened results or repeated failure.

## Accessibility
- Each row is a `listitem` whose name includes title, status and next run; auto-pause reasons are in text, not only an icon or colour.
- After Run now, announce "Running" and then "Delivered to <destination>".

## House rules
- Real `<button>`/`<input>` elements, full keyboard use, visible focus; announce status changes with `aria-live="polite"`; respect `prefers-reduced-motion`.
- All user-facing text comes from props (or a `labels` prop) so it can be translated.
- One consistent accent marks anything the AI made or is doing; amber means "needs a person", red means risky, green means done. Never rely on colour alone.
- Fits a 360 px wide container without horizontal scrolling.

## Acceptance checks
- [ ] Every state above can be rendered by passing props, with no internal fake data.
- [ ] Every interaction and edge case above behaves as described.
- [ ] Nothing in "Must avoid" can happen with this component.

## Deliver
1. The component file with its exported prop types.
2. A short usage example that renders each state.

Download this prompt as Markdown, or every prompt as JSON for an agent.

Critique checklist for dark matter

0 of 4 checked

These questions apply to every pattern in this family. Ticks are saved in this browser.

Can the user find out what ran without being told?
Is background work reversible after the fact?
Do failures surface, or die silently in a log?
Is there one place that lists everything running on the user's behalf?