It’s morning. I’m making coffee, and I’m about to head to the bakery around the corner before I sit down for work. Before I leave, I tell ChatGPT to search the web for the latest deals on the running shoes I’ve been wanting to buy. I instruct it to give me options but not to purchase anything. Then, I ask it to find me travel options for my next marathon race. I want to fly as early as possible in the morning and stay at a hotel as close as possible to the venue. Finally, I instruct ChatGPT to find me restaurant options for a family dinner I’ve been postponing. It’s going to be a party of 10 with different culinary preferences, and it’s going to be complicated.
I’m not even out the door before I give the AI another command. With the next-gen iPhone launch approaching, I want an in-depth look at how we got to the latest design changes. I want a report on iPhone sales around the world with trends for key markets, looking for correlations between big iPhone redesigns and uptick in sales.
Now I’m out the door in search of that tasty croissant I’m telling myself I’ll burn burn off with my next run, knowing that the voice instructions I gave the AI will yield results within half an hour. No point wasting time around the apartment before I actually need to sit in front of the computer to start work. I know that ChatGPT will ping my phone as each of the tasks I gave it is completed.
This wasn’t a real morning of mine, and it won’t be for quite some time. But with ChatGPT Operator and Deep Research now here, we’re getting close to a future where AI agents will help us get things done faster and free up time for other activities. Yet, in spite of these initial launches, I’m already wondering whether the AI will get it right and whether ChatGPT will be smart enough to pass instructions between AI agents.
OpenAI launched Operator a few weeks ago, with the AI agent available to ChatGPT Pro users in the US. As a ChatGPT Plus subscriber in the EU, I don’t have access to it. But in my example above, I would hopefully use ChatGPT Advanced Voice Mode to kickstart Operator actions.
Operator will search for shoes in a virtual computer in one of ChatGPT’s tabs. The same AI agent will then browse the web in a different virtual machine to look for travel options, already knowing my preferences.
I would expect a combination of Operator and Tasks to handle my dinner reservations. Operator would open another tab to look for restaurants that match my needs, and Tasks would save reminders as the reservation is handled.
Operator isn’t good enough to give me the iPhone report I need, so that’s where ChatGPT will involve Deep Research, this time in a different ChatGPT window, to perform the lengthier search.
Deep Research is the AI agent ChatGPT released on Sunday. Again, it’s not available to ChatGPT Plus users, and it’s not available in the EU. The research tool can get you more in-depth reports on virtually any topic, whether you’re conducting work-related research or just looking for the right running shoes.
Unlike Operator, Deep Research will not assist with web actions like buying items or making reservations. It also won’t run on a virtual machine to browse the web.
Operator and Deep Research are separate ChatGPT functionalities right now. You invoke them differently, and I don’t even know whether Advanced Voice Mode is supported or whether I can use voice to run either AI agent. I’d imagine the feature is coming in the future.
Voice integration will be key to making these AI agents work for you. I don’t want to go to the ChatGPT app and figure out the right button for Operator, Deep Research, and any future AI agent OpenAI might come up with. I want to tell the AI what I need by voice, and then ChatGPT will figure out which of its tools it needs to use to get the job done.
ChatGPT will know what task needs Deep Research and where to use Operator instead. Since the AI is available on Mac and iPhone, I can check the progress and tweak prompts while out of the house. ChatGPT will notify me, or just talk to me if I’m already wearing AirPods, to issue updates on the tasks, ask follow-up questions, or just tell me that everything is done.
By the time I get back home, a bag of croissants in my hand and music in my ears, I’d be a few steps ahead of my day. Research for some of the work I had planned would be done. Some of my personal chores would be handled. All I’d have to do is decide whether to buy the shoes, confirm the travel plans, and then notify my family about the dinner plans.
It doesn’t sound like a lot, but ChatGPT and similar AIs with agentic powers will save me minutes every day. In turn, I’ll use that time for me. Rinse and repeat throughout the day, and I’d get more minutes for myself.
It’s not immediately clear, but AI agents can become the true assistants we’ve been waiting for since Apple first unveiled Siri many years ago.
Give the AI the ability to also control computers and devices near you, and computing will be similar to what we saw in Star Trek growing up or to Iron Man (Robert Downey Jr.) talking to his computers in Marvel movies. We’ll use a conversational tone to issue more and more complex tasks to the AI and control every type of computer where that AI lives.
It’s all starting now, with OpenAI laying the foundation of that computing experience. Advanced Voice Mode might not be available for AI agents, and ChatGPT might not know on its own to pass a task to Operator or Deep Research. But these various tools will surely be bridged in the not-too-distant future.
My scenario involves ChatGPT because OpenAI is leading the pack here. But I expect Google, Microsoft, Meta, Apple, Anthropic, DeepSeek, and a slew of other tech and AI firms to conceive similar AI agent/assistant experiences and ecosystems in the near future.
The post ChatGPT’s AI agents are here, and this is how I want to use them appeared first on BGR.
