How to automate tasks in ChatGPT in 2026: Scheduled Tasks, Work, Codex and when to go further
ChatGPT can now run work on a schedule, react to events and finish multi-step tasks. Here is how each piece works and where it stops.
You can automate ChatGPT in three ways. Scheduled Tasks run a prompt once, on a schedule or on an event such as a new Gmail message, Slack message or GitHub pull request. ChatGPT Work handles longer multi-step jobs and produces finished documents, spreadsheets and presentations. Codex handles coding tasks in cloud sandboxes. All three share one usage allowance, so heavy automation can use up your assistant time.
Scheduled Tasks: run work without you
A scheduled task runs a prompt once, repeats on a schedule or trigger, or monitors for changes. OpenAI's documentation says scheduled tasks "run unattended". Time-based schedules can be minute-based intervals, daily or weekly, with advanced recurrence rules for finer control.
On eligible plans, tasks can also respond to app events: new Gmail messages (optionally filtered by sender or subject), new Slack messages in chosen channels, and GitHub pull request activity. Event triggers work in ChatGPT on the web and mobile, not in the desktop app or the IDE extension.
ChatGPT Work: longer jobs with finished output
OpenAI describes Work as an agent designed for longer, multi-step work and finished deliverables. On web and mobile it runs in the cloud, and it can research, work across connected apps and files, and create documents, spreadsheets, presentations, reports and Sites. It has a cloud browser that can operate supported websites and hand control back to you for sign-in.
Work is included with Plus, Pro and Business plans, and not with Free or Go.
Codex: code in cloud sandboxes
Codex runs coding tasks in the cloud and locally. OpenAI states that Work and Codex share the same usage allowance and credit pool, and that "local messages and cloud chats share your plan's usage allowance" with weekly limits that may also apply.
Where ChatGPT automation runs out
- One shared pool. A large Codex build and a long Work task draw on the same allowance as your everyday chat.
- OpenAI models only. We found no way to choose another vendor's model inside ChatGPT.
- Website blocks. Some sites block automated browsing, and not every transaction can be automated.
- Consumer data default. Content on consumer plans is used to improve models unless you turn that off in Data controls.
When to add another tool
| Need | Option |
|---|---|
| Connect thousands of apps with simple triggers | A workflow tool such as Zapier or n8n |
| Use Claude, Gemini or Grok as well as OpenAI, on a separate allowance | AutoAGI |
| Run cloud coding agents beyond your ChatGPT allowance | Claude Code and Codex in the cloud |
| Build agents you fully control and can self-host | AutoGPT or another framework |
| Classic prompt-chaining workflows inside a ChatGPT Plus tab | AGI Layer for ChatGPT |
Questions
Can ChatGPT run tasks on a schedule?
Can ChatGPT trigger a task from Gmail or Slack?
Which ChatGPT plans include ChatGPT Work?
Is ChatGPT automation free?
What is the difference between ChatGPT and AGI Layer?
Related comparisons and guides
AutoAGI vs ChatGPT: where an autonomous workspace beats an assistant, and where it does not
AutoAGI vs ChatGPT Work. ChatGPT Plus is $20 and runs OpenAI models on a shared allowance with Codex; AutoAGI is $29 to $200 with Claude, OpenAI, Gemini, Grok and weekly usage in dollars.
GuideAutonomous AI agents: what they are, how to choose one and what they cost
A plain guide to autonomous AI agents: a clear definition, how they differ from chat assistants, the criteria that matter when choosing one, how usage is metered and which products fit which job.