ChatGPT Agent can browse websites, run code, and create documents. This tutorial explains how to start a task, follow its progress, and review the result.
What is ChatGPT Agent?
ChatGPT Agent uses a virtual computer to browse websites, fill forms, run code, analyze data, and create slides or spreadsheets. It brings together features from Deep Research and Operator.
For the later Work interface and how it relates to coding tasks, see my ChatGPT Work, Chat, and Codex guide.
You can follow the agent’s progress as it researches and uses tools. It asks for permission at certain steps.
For example, you can ask it to browse flight sites, compare prices, and return a table of options for you to review.
Think of it this way:
- ChatGPT. You ask, “How do I research my competitors?” It gives you a step-by-step guide.
- ChatGPT Agent. You say, “Research my top 5 competitors and put the findings in a spreadsheet.” It then opens a browser, performs the search, analyzes the data, creates a spreadsheet, and hands you the finished file.
The agent uses a virtual computer to browse, execute code, and create documents. Its progress view shows the tools it uses and the results it finds.
How to Use ChatGPT Agent Mode
Currently, it’s not available for free users. It is available to ChatGPT Pro, Plus, and Team subscribers. Will roll out to Enterprise and Education soon. Not available in the European Economic Area (EEA) or Switzerland yet.
Open any ChatGPT website and open a chat window. Then, click the Tools dropdown and select Agent Mode, or simply type /agent into the chat box.
Write a clear and detailed prompt that describes exactly what you want. Specify desired outcome (e.g., “create a 5‑slide deck summarizing competitor strengths”). Include constraints (e.g., budget, dates, file format) and add context, like travel dates or project specifications. The clearer the instructions, the better your results.
Once prompted, the agent launches its virtual desktop and tools. It may ask follow-up questions if clarification is needed on dates, preferences, or formats.
You can always stop the agent and adjust the prompt or event, or take over the browser or terminal window.
If you want to know how to do that, please watch my video tutorial:
Watch on YouTube: ChatGPT Agent Mode
ChatGPT Agent Mode Limitation
Tasks typically take between 5 and 30 minutes, depending on their complexity.
Monthly usage caps:
- Pro: ~400 agent messages/month
- Plus: ~40 messages/month
- Team: ~30 credits/month
You can see the remaining number of prompts near the Agent Mode menu in the tooltip.
5 top real‑world use cases for ChatGPT Agent mode
- Market & Competitor Research. Task: “Analyze top 5 competitors and compile a spreadsheet report.” The agent can browse websites, extract information, compare features and prices, and produce a spreadsheet. Check the sources and figures before using it.
- Trip Planning & Booking Assistance. Task: “Compare flights, hotels, and an itinerary under a specified budget.” The agent can browse booking sites and return a table of options. Verify prices and travel requirements, and review any booking before approving it.
- Presentation Generation (Slide Decks). Task: “Create a slide presentation on a topic such as market size or global statistics.”
Why: The Agent gathers data, writes and runs code to generate charts, inserts images, and builds fully editable slide decks (e.g. in Google Slides or PowerPoint), ready for your finishing touches. - Email & Message Automation. Task: “Draft responses to these messages.” The agent can prepare drafts using connectors such as Gmail or Slack. Review the text and recipients before approving a send.
- Data Processing & Spreadsheet Management. Task: “Update or clean a spreadsheet from raw data.” The Agent works with spreadsheets: updating contact lists, checking websites, formatting data, or summarizing project status. Suitable for lead generation, reporting, or internal document creation.
These are just the minimum that you can do with the agent. In my YouTube video, I show three real examples.
Conclusion
Be careful with the information and account access you provide. Review the agent’s work and the permissions it requests before approving an action.
Start with a task whose result you can check, such as a short research table or a draft document.
Please give it a shot and share with me your feedback in the comments below!
Cheers! ;)
