ChatGPT Now Sets Its Own Reminders — OpenAI's Scheduled Tasks Turn the Chatbot into a Proactive Assistant

For years, ChatGPT has been reactive: you ask, it answers. That changed on June 17, 2026, when OpenAI launched scheduled tasks across all paid tiers, turning the chatbot into something that can actually work without being summoned.

The premise is straightforward. Users can now tell ChatGPT to send a reminder at a specific time, run a recurring weekly task, or silently monitor a topic on the web and report back only when something worth noting happens. A dedicated Scheduled page in the sidebar lists every active task, shows when it runs next, and lets you pause, edit, or delete it without digging through old chats.

Scheduling is deliberately flexible. You can pin a task to an exact time or describe a window — "every morning," "Tuesday afternoons" — and ChatGPT figures out the slot. Monitoring tasks are where the feature gets genuinely interesting: they can search the web, check connected apps, and surface only meaningful changes rather than dumping a daily digest of noise. Tasks can fire at most once per hour, and unused ones pause automatically if left idle for long enough.

Limits scale with plan tier. Go users get three simultaneous active tasks; Plus users get five; Business and Education accounts get ten; Pro and Enterprise users top out at fifteen. Across all tiers the rollout covers web, iOS, Android, and macOS, with Windows support on the way.

The launch also retired Pulse, OpenAI's earlier proactive-summary experiment. Pulse ran daily briefings on tracked topics, but users never had much control over timing or triggers. Scheduled tasks replaces that with a more general, user-directed model — you define what runs, when, and what counts as a notification-worthy event. Pro users have a 14-day window to recreate their Pulse setups before it goes dark.

The bigger picture is that this is part of a broader push by every major AI lab to make models less passive. OpenAI's product cadence has been moving toward agentic workflows all year — GPT-5.5 as the default model, scheduled tasks, connected-app monitoring. The gap between "AI you have to ask" and "AI that works on your behalf" is closing fast, and scheduled tasks is the clearest consumer-facing sign of that yet.

What it won't do, at least for now, is take action autonomously. Tasks report and remind; they don't draft emails, submit forms, or trigger external services on your behalf. That distinction is likely deliberate — a way to build trust with the feature before expanding its reach. But it's a meaningful first step toward a ChatGPT that shows up in your workflow instead of waiting to be opened.