ChatGPT's New Task Scheduling Feature | Baby Step to the Agentic Era?

OpenAI’s new “Tasks” feature inside ChatGPT lets users schedule reminders and actions to run automatically at specific times—up to 10 active tasks—marking a small but concrete step toward more agent-like behavior. The practical promise is straightforward: set an instruction once (for example, “update me with daily news at 4 p.m.”), and ChatGPT will later trigger the work on a timer and deliver the result in the same chat thread, with optional desktop notifications.

Early examples range from playful to work-adjacent. A daily “knock knock” joke at 9:00 p.m. and a morning fantasy basketball recap show how tasks can reduce repeated prompting. More useful concepts appear in the “creator economy” direction: scheduled summaries tailored to a domain, weekly meal plans, and recurring reminders tied to personal schedules like language practice or birthdays. The feature is rolling out as a beta to ChatGPT Plus, Team, and Pro subscribers over the next few days, and it’s tied to a specific model variant—“GPT 4o” with scheduled tasks—rather than being available across all ChatGPT models.

Hands-on testing highlights both what works and what’s missing. Tasks can be created through the ChatGPT interface, appear with a schedule UI, and run by “kicking off” API activity at the chosen time. The scheduled output can include web retrieval and other limited capabilities, but the feature’s tool access is constrained: file uploading is unavailable, and the model’s ability to generate images via DALL·E appears unreliable in at least one test (an image generation attempt failed before succeeding on retry). Notifications also appear inconsistent on mobile—notifications may be allowed in settings, yet the tester reported not receiving them in the ChatGPT iPhone app.

A key limitation is that tasks don’t yet behave like a fully autonomous agent that can act across the user’s life. Meal planning, for instance, would require ongoing context about preferences and inventory—information that would need to be provided or updated, which can erode the “set it and forget it” value. The same gap shows up in broader ambitions: the tester wants tasks that can connect to email, calendars, Google Drive, or even computer workflows (e.g., checking YouTube comments, scanning inboxes, or using browser automation). Other automation platforms like Zapier and Lindy already support scheduled triggers and API connections, but Tasks currently looks closer to “scheduled prompting” than “delegated execution.”

Still, the direction matters. Tasks can access “memories” managed in ChatGPT settings, and completed tasks may send email updates tied to the user’s ChatGPT account—though viewing them may require clicking through to the web app. The overall takeaway is that Tasks is a foundational capability—timed execution inside ChatGPT—but its usefulness depends on future expansion: deeper tool access, reliable multimodal actions, and tighter integration with external services. If OpenAI extends Tasks to connect to the user’s accounts and devices, it could become a genuine building block for the agentic era; for now, it’s an early, somewhat underwhelming step.