Master OpenAI: Turn ChatGPT Into Your Productivity Engine / Creating and Customising
ChatGPT Work and Current Agent Features
About 15 minutes
OpenAI's agent features have changed quickly. This lesson covers what is current as of 18 November 2026, and how to use it safely. Always check the help centre for the latest state.
What changed
OpenAI's help centre now says the earlier "ChatGPT agent" feature is no longer available. For longer, multi-step tasks and finished deliverables, it points users to ChatGPT Work, and to the cloud browser for supported browser tasks. If you read an older guide that tells you to pick "Agent mode" from the tools menu or type /agent, that guidance is out of date.
What ChatGPT Work does
According to OpenAI's help pages and release notes, Work is designed for longer, more involved tasks. It can research and analyse information, work across connected apps and files, and create finished documents, spreadsheets, presentations and reports. It has a built-in browser for gathering information from websites and working with supported web tools.
Who has it
OpenAI's help centre says Work is for paid plans, not Free or Go, and has been rolling out gradually. Usage is included with Plus and Pro and with Business seats, with limits. The help centre also notes Work is not available in projects that use project-only memory. If you do not see it, your account may not have access yet.
Step 1: Pick a suitable task
Good Work tasks:
- Have a clear finished output (a report, spreadsheet or slide deck).
- Need several steps: gather, analyse, write.
- Would take you an hour or more by hand.
Poor Work tasks:
- Quick questions. Use a normal chat.
- Anything involving payments, passwords or sending messages to clients without your review.
Step 2: Write a task brief
Goal: Produce a 6-slide deck comparing three accounting tools for a 5-person agency.
Inputs: Use the attached spreadsheet of our current costs.
Research: Use official pricing pages only. Record the date checked for each price.
Output: Slides with a comparison table, pros and cons, and a recommendation slide.
Stop and ask me before signing in to any site or submitting any form.
The last line matters. It keeps you in control of consequential steps.
Step 3: Supervise
- Watch the progress where shown.
- Interrupt if it drifts or uses poor sources.
- If it needs to sign in to a site, follow OpenAI's guidance: do not type passwords into chat messages. Use the takeover or sign-in method the product provides, or do that step yourself.
Step 4: Review the output
Treat Work outputs like a junior colleague's first draft:
- Check every figure against its source.
- Read every sentence that will go to a client.
- Open the file it created and check the formatting.
Scheduled tasks
ChatGPT also supports scheduled tasks, which run a prompt at a set time, such as a weekly summary. Availability depends on your plan; see the "Scheduled tasks in ChatGPT" help article.
Every Monday at 8am, search for news about [my industry] in [my country]
from the past 7 days and give me 5 bullet points with links.
Safety rules for any agent feature
- Never share passwords or card details in chat.
- Require approval before purchases, form submissions or sending messages.
- Start with read-only tasks: research and drafting.
- Keep a human check on anything a customer will see.
Common mistakes
- Following old tutorials. Agent features have been renamed and replaced. Check the help centre date.
- Vague goals. Give a clear finished output.
- Leaving it unsupervised on sensitive sites. Stay in control of sign-ins and submissions.
- Using Work for small jobs. A normal chat is quicker and uses fewer limits.
Exercise
- Check whether ChatGPT Work is in your account.
- If yes, write a task brief using the template and run it on a research-and-report task. If no, run the same brief as a deep research task.
- Verify three facts in the output.