ChatGPT Agent Mode - An Important Breakthrough

In the AI world, 2025 is clearly the year of the agent. There is a lot of hype around AI agents and agentic AI. As is usually the case, the hype is overblown (that’s why it’s hype). But, in this case, there’s a thread of reality in the hype.

The most recent bit of hype came from OpenAI’s release of agent mode, which sounds like the name of a comic book character or maybe a secondary character in Get Smart. Agent mode allows ChatGPT to take independent action on its own virtual computer. Here’s an excerpt from OpenAI’s announcement of its new agent.

ChatGPT can now do work for you using its own computer, handling complex tasks from start to finish.

You can now ask ChatGPT to handle requests like “look at my calendar and brief me on upcoming client meetings based on recent news,” “plan and buy ingredients to make Japanese breakfast for four,” and “analyze three competitors and create a slide deck.” ChatGPT will intelligently navigate websites, filter results, prompt you to log in securely when needed, run code, conduct analysis, and even deliver editable slideshows and spreadsheets that summarize its findings.

At the core of this new capability is a unified agentic system. It brings together three strengths of earlier breakthroughs: Operator’s⁠ ability to interact with websites, deep research’s⁠ skill in synthesizing information, and ChatGPT’s intelligence and conversational fluency.

ChatGPT carries out these tasks using its own virtual computer, fluidly shifting between reasoning and action to handle complex workflows from start to finish, all based on your instructions.

Agent mode is pretty remarkable, although I’m not sure it’s a true AI agent. (More on the distinction later.) To use Agent Mode, you prompt ChatGPT as normal, but select Agent mode. (If you don’t see Agent mode, you probably don’t have access yet. It’s available to Pro users (the $200 per month plan) and is being rolled out to other paid users over the next few weeks.)

Once you put in the prompt, ChatGPT gets to work. You can actually watch Agent mode do its work. It shows the “screen” of the virtual computer, overlaid with its reasoning process. This is pretty awesome and oddly fascinating. It’s hard not to watch it do its work.

Agent mode is persistent. As you can see above, it tried to open an IBM site, but was thwarted. That won’t stop AGENT MODE! It just tries a different path to get the desired data or information.

Some tests of agent mode

Just like with humans, the difficulty of the task determines how long it takes agent mode to complete its work. In one of my early tests, I asked agent mode to analyze the back catalog of episodes of a podcast I do, Live Well and Flourish. There are over 100 episodes of the podcast, so ChatGPT had quite a bit of work to do. In 28 minutes, it produced a nicely formatted analysis of the themes in Live Well and Flourish, along with suggestions for future episodes and an analysis of similar podcasts. It would have taken a human assistant several hours to create a similar report.

In another test, I asked it to create a schedule of my sessions for an upcoming conference. This was a pretty simple task that took only two minutes. Doing the same thing myself took about 5 minutes. Not a huge savings, but better than nothing.

I also asked agent mode to create a slide deck for a presentation I’m giving at the same conference. I gave ChatGPT the paper that I’ll be presenting. It took ChatGPT 39 minutes to create a decent, but not great slide deck. Much of this time was spent creating or finding images. Although agent mode completed the task successfully, the results were OK at best. In contrast, Beautiful.ai created a much better looking slide deck in two or three minutes. That being said, I still consider this test a success. If I didn’t have Beautiful.ai, I would have used the ChatGPT created slide as a starting point for my presentation.

In a final test, I asked agent mode to create an overview of AI agents versus agentic AI. It worked for 13 minutes and produced a pretty good report, which you can access here. This is a muddled area. Frankly, I’m not sure there are good, widely agreed upon distinctions between AI agents and agentic AI. The report matched the current thinking which considers agentic AI as collections of self-managed agents that work together to achieve some goal.

Overall, agent mode is impressive. It can produce credible results autonomously, which is a significant step forward in mainstream AI capabilities. The AI agent hype still outstrips the reality, but the reality is increasingly capable.

Agent mode and higher ed

Tools like agent mode have significant potential benefits for higher ed professionals. One of the banes of my existence is updating important dates in a learning management system. I often teach the same class three terms in a row. So, I roll over the previous term’s learning management system (LMS) site to the new term. Of course, all of the dates need to be changed. This class has a ton of online activities, so there are a lot of dates to update. I ALWAYS miss a few. Students panic when their assignment is marked as late, and I can’t blame them since the error is mine. Someone emails, I change the date and all is well. There’s absolutely no reason that an AI agent couldn’t go through and change the dates for me, or at least check to see if I missed any changes. Agent mode will let you take control to do things like log into a system, so this is feasible right now, although I haven’t tried it yet since I’m on sabbatical in the coming year!

The possibilities for tools like agent mode are almost endless. They have the potential to free us from a plethora of mundane tasks, freeing us to do things that are more important and rewarding. Sure, agents are overhyped at the moment, but remember, sometimes the hype turns out to be an understatement in the long run.

---

Want to continue this conversation? I'd love to hear your thoughts on how you're using AI. Drop me a line at Craig@AIGoesToCollege.com. Be sure to check out the AI Goes to College podcast, which I co-host with Dr. Robert E. Crossler. It's available at https://www.aigoestocollege.com/follow.

Looking for practical guidance on AI in higher education? I offer engaging workshops and talks—both remotely and in person—on using AI to enhance learning while preserving academic integrity. Email me to discuss bringing these insights to your institution, or feel free to share my contact information with your professional development team.