OpenAI’s release of GPT-5 has dominated recent headlines, but arguably the more important news is OpenAI's launch of ChatGPT Agent Mode, a new mode that makes use of AI agents, which many see as the next generation of AI.
Unlike a traditional chatbot, including ChatGPT, AI agents can perform tasks such as check and send emails, create and complete spreadsheets, run code, check hotel reservations, order ingredients, and more. In other words, AI agents are designed to work as a true AI assistants.
OpenAI released ChatGPT Agent Mode in July but the news was quickly overshadowed by the release of GPT-5 in early August.
Of course, the release of a tool such as this will have major implications for educators and our students. I've written about how GPT-5 didn’t seem like a major game changer for AI in education, but agent Mode is different. This feels like a major new capability for AI that will have a significant impact on how we integrate AI into our classrooms and guard against inappropriate AI use by students.
After experimenting with AI agents for less than a week, it’s already saved me more time than any other AI tools, and has left me more worried about a whole new class of potential AI misuse.
Here’s more detail on everything you need to know about ChatGPT’s Agent Mode, which is available for subscribers to Pro, Plus, Team, Enterprise, and Edu ChatGPT plans.
What is ChatGPT Agent Mode?
Agent Mode combines two existing ChatGPT features: Operator and Deep Research. In OpenAI’s words, Agent Mode brings together “Operator’s ability to interact with websites," and "Deep Research’s skill in synthesizing information.”
In more concrete terms, Agent Mode gives ChatGPT the ability to open a virtual desktop computer, browse the internet, and complete work much the same way a human assistant might. This gives the tool the ability to do more than research and answer questions for you; it can perform specific tasks.
For instance, OpenAI boasts Agent Mode can handle requests such as:
- Look at your calendar and brief you on upcoming client meetings based on recent news
- Plan and buy ingredients to make Japanese breakfast for four
- Analyze three competitors and create a slide deck.
It doesn’t take a big imagination to think of how this can be useful for educators. For administrative tasks, educators can have it help organize emails, create spreadsheets, build more robust lesson plans, and more.
For students, it's markedly different than other AI tools in that it does not necessarily lend itself well to a tutoring role. That said, it can still help them with their research-related tasks and cut down on busywork, much as it can do for teachers.
It also makes cheating with AI that much easier, however. Students could theoretically use it to respond to teacher questions about their AI-generated essays to complete worksheets and online quizzes, and probably much more than I can even think of as students tend to be at their most creative when it comes to this type of thing.
How Do You Access Agent Mode?
Agent Mode is available to Pro, Plus, Team, Enterprise, and Edu. These plans start at $20 for the Plus plan per month and increase significantly from there.
I’ve found my Plus subscription to be adequate for my needs, but you’re limited in how much you can use some advanced features, including Agent Mode, each month. Additionally, you don’t always get access to the most advanced models.
Right now, GPT Pro is only available to Pro subscribers. A Pro subscription is $200 per month.
What Are Some Specific Task Educators Can Use Agent Mode For?
I found agent mode to be incredibly helpful as a research assistant. I sometimes write about obscure topics and find myself searching large documents for little bits of specific info. ChatGPT Agent Mode makes this type of work effortless and can search for specific mentions within a document and more.
I also recently had Agent Mode search for every journalism professor job listing in New England and the metro area and found it provided a great overview of all the current jobs—and though a quick Google search probably would have turned up similar results, it also put it all into a spreadsheet that would allow me to track submissions.
I also have used it to create a mock PowerPoint presentation, and it was a huge improvement from past experiences I’ve had attempting to use AI for help creating a presentation. In the spring when I tried to use AI to create a presentation, I got visually unappealing documents with text that had little depth and lots of hallucinations. Agent Mode was much better. It worked on my presentation for an hour and returned a visually appealing presentation with links to relevant sources that was as good, and arguably a little better than, I could do on my own.
I didn’t feel comfortable giving ChatGPT Agent Mode access to my emails from a security perspective, and because using AI to respond to emails feels incredibly inauthentic. That said, the efficacy of Agent Mode has made me a little paranoid, and now I worry that I’m actually writing back and forth with ChatGPT during some quick email exchanges.
On the other side of the equation, as mentioned previously, Agent Mode does seem like it will lend itself to cheating. I teach online, and AI submissions are already a significant problem. Even so, I’ve never felt like a student was using AI to actually log into the course for them—that’s conceivable now, and a bit scary.
Bottom Line
From an administrative perspective, ChatGPT's Agent Mode is a really helpful tool. I’m going to continue to think about ways I can use it to help with spreadsheet-related tasks and as a research assistant.
Even though I worry about the spread of this technology in the classroom and beyond, I’m more excited about this feature than I’ve been about most recent releases from OpenAI and other AI companies.
Ultimately, ChatGPT Agent Mode is one of the more practical AI tools I’ve come across.
