ChatGPT Scheduled Tasks Officially Launched: Precise Scheduling, Periodic Execution, Pulse Retires in Two Weeks

On June 18, OpenAI began gradually rolling out a long-awaited feature to ChatGPT users — the brand-new Scheduled Tasks. Starting today, Plus, Pro, Business, and Enterprise users will begin seeing a new "Scheduled Tasks" entry in the sidebar. At the same time, OpenAI has given the old Pulse feature a death sentence: it will officially shut down in 14 days, and all active push tasks will be handed over to the new system.

This may not sound like a huge deal, but looking at OpenAI’s recent product pace, the signal is clear: ChatGPT is shifting from a simple “ask-and-answer” tool toward becoming a persistent assistant that can take notes, monitor tasks, and execute on schedule.

One-Sentence Summary of What It Can Do

Scheduled Tasks are quite straightforward — they let ChatGPT work on your behalf according to a plan when you’re not around. OpenAI officially lists three typical scenarios: sending reminders, running periodic work, and monitoring changes.

Here are some examples developers might find useful:

• Every morning at 8:00, fetch the Hacker News front page and new cs.AI papers from arXiv, summarize, and send to you.
• Every Monday morning, check the Release Notes of certain GitHub repositories and notify you if there's an update.
• Monitor a specific exchange rate / stock price / API availability, and alert when a trigger condition is met.
• On the first day of the month, automatically organize last month’s Stripe billing data structure.

Previously, you needed to build a cron job + script + notification webhook to handle these tasks. Now, you can just hand them to ChatGPT. Of course, whether it can fully replace scripts is something we’ll discuss below.

What Changed in This Update

Compared to the initial Tasks release in January, this version can be broken down into four major changes:

1. Dedicated Task Management Page

The sidebar now has a dedicated “Scheduled Tasks” page where you can see all active, paused tasks, and their next triggers at a glance. You can pause, resume, edit, and delete directly here.

It sounds basic, but veteran users know that previously Tasks was buried deep inside settings, making management a pain — once you had several, you just wouldn’t bother checking them. Elevating it to a primary menu item shows OpenAI intends this feature to be used frequently.

2. More Flexible Scheduling Granularity

You can specify both exact times (e.g., Wednesday at 15:30) or broad time ranges (morning, noon, evening).

The latter may seem trivial, but it’s actually quite user-friendly. Many tasks don’t need to run exactly at 8:00 vs. 8:15 — giving the system a time window lets it execute when load is lower, reducing peak-time congestion and avoiding mass simultaneous notifications.

3. Monitoring Tasks Support “Worth Notifying” Filter

This is perhaps the most interesting update. Monitoring tasks can search online and read data changes from connected apps (Connectors), but OpenAI explicitly states: notifications are sent only when there’s something worth informing you about.

In other words, the model itself applies a layer of “importance assessment.” If you ask it to check a repository’s issues every hour, it won’t spam you with “no change” updates — it’ll alert only when there’s a genuine change.

This is the right direction — the biggest UX killer for active-push agents is noise. Pulse was often criticized for “useless morning briefings.” OpenAI clearly learned this lesson and gave decision-making power back to the model.

4. Speed and Stability

The official note is: “All tasks have improved speed and stability,” which is boilerplate for any update. But considering Tasks have often failed over the past six months (missed triggers, hours-long delays), if this is truly fixed, it would be a qualitative improvement.

Key Limitations — A Reality Check

OpenAI’s fine print includes two hard limits developers should know:

1. Tasks can run at most once per hour. If you want second-level or minute-level monitoring, forget it — just use your own cron.
2. Long-running unattended tasks may be automatically paused. This is important — it means if you set a monthly task but don’t open ChatGPT for several months, it might silently stop. This is a significant limitation for true automated workflows.

These two restrictions indicate OpenAI sees Tasks as an enhanced personal assistant, not a replacement for automation platforms like Zapier or n8n. For unattended production-grade workflows, you still need API + custom scheduler.

Relationship with Pulse: Complete Takeover

Pulse was OpenAI’s active-push feature introduced for Pro users last year, intended to send personalized morning briefings. The idea was good, but the problem was clear — you couldn’t control the push content, and the model’s picks were hit-or-miss.

Now with the new Tasks, Pulse will shut down in 14 days. OpenAI’s logic is clear: rather than guessing what you want, let you define it yourself. One task, one sentence — what to monitor, when to push, and to what extent — all in your control.

From a product philosophy standpoint, this is a retreat from “AI recommendation” to “user orchestration.” It may sound like a step back, but in active-push scenarios that rely heavily on personalization, letting the user control the trigger logic is more sustainable than having the AI guess.

Competitive Comparison: Claude, Gemini

Talking about scheduled tasks inevitably leads to competitors:

• Anthropic Claude: No native scheduled execution yet, relying on third-party MCP Server for scheduling. Projects provides persistent context but not active triggers.
• Google Gemini: Well-integrated with Workspace ecosystem — Gmail, Calendar — but still lacks a standalone “scheduled tasks” product.
• Microsoft Copilot: Has full scheduling in Power Automate, but that’s enterprise automation, not lightweight tasks inside a chat interface.

So ChatGPT’s move means it’s the first to make scheduling a first-class citizen in a consumer-grade chat product — a first-mover advantage, especially for developers and creators not yet using professional automation tools.

Developer Perspective: Where to Use, Where Not to Use

Some practical guidelines:

Tasks are good for:

• Information aggregation and summaries (daily news, papers, industry updates)
• Personal reminders (birthdays, renewals, reviews)
• Low-frequency monitoring (daily/weekly checks of a webpage or data)
• Content creation assistance (weekly brainstorm topic generation)

Avoid using Tasks for:

• Second/minute-level monitoring (hourly limit)
• Long-term unattended pipelines (may be auto-paused)
• Production tasks needing deterministic output and audit logs (model behavior is nondeterministic)
• Critical operations (like automatic ordering or transferring funds) — don’t risk it

A reasonable hybrid architecture: Tasks handle human-interaction-layer active reminders, cron handles production-layer hard scheduling, connected via API.

Other Client Changes This Week

A quick note on OpenAI’s recent client-side progress — showing they’re systematically enhancing desktop experiences:

• Image processing improvements in ChatGPT client
• Mac Codex adds Dock icon customization and quota reset cache saving
• Codex launches Computer Use capability in the EU

Taken together, OpenAI’s product strategy is clear: ChatGPT as a persistent assistant, Codex as a desktop-level agent, developing in parallel.

Conclusion

This Tasks update isn’t technically groundbreaking — it’s essentially solidifying “scheduling + triggers + notifications” and adding a model-driven “worth notifying” judgment layer. But in ChatGPT’s product evolution, it’s a key step from passive response toward active companionship.

By the way, when testing new models and features across multiple platforms, an aggregation platform like OpenAI Hub can save effort — one key can call GPT, Claude, Gemini, DeepSeek, with OpenAI-format compatibility and direct access from China. If you’re planning mixed scheduling like “ChatGPT compiles summaries every morning, Claude does deep review on weekends,” you’ll configure fewer SDKs.

As for Pulse, farewell in 14 days — a product retiring with dignity after being fully replaced.

References

• IT Home: OpenAI Launches New Scheduled Tasks Feature for ChatGPT, Rolling Out from Today: Details of the update and rollout scope.
• Zhihu: ChatGPT Unlocks Autonomous Execution for the First Time, OpenAI Steps into L3: Product analysis of initial Tasks release, useful background for this update.