Sidebar Sidebar
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.

OpenAI Launches ChatGPT Agent That Can Complete Tasks For You

MacRumors

MacRumors

macrumors bot
Original poster
Apr 12, 2001
68,546
39,401


OpenAI today launched ChatGPT agent, a new agentic model that is able to think proactively and complete computer-based tasks on the user's behalf.

chatgpt-logo.jpg

The ChatGPT agent is in the same family as o3. It combines several existing ChatGPT features, and it can do things like research and generate reports, execute code using Terminal, generate slides and spreadsheets, and connect to external data sources and applications.

OpenAI gives several examples of how ChatGPT agent can be used:
  • 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.
  • Analyze three competitors and create a slide deck.
The ChatGPT agent uses its own virtual computer, and it will navigate websites, filter results, prompt users to log into websites when needed, and deliver summaries of its findings. It is designed to seek permission before taking any "actions of consequence," and OpenAI says that users can interrupt tasks to add extra instructions, and stop tasks at any point.

ChatGPT agent is rolling out starting today for Pro, Plus, and Team users. Just select "agent mode" from the dropdown menu in the composer during a conversation. ChatGPT users are able to transition between conversations and action requests within the same chat.

Pro users will get access by the end of today, while Plus and Team users will get access over the next few days. OpenAI plans to add the functionality for Enterprise and Education users in the coming weeks. Pro users have access to 400 messages per month, and other paid users will get 40 messages monthly with additional usage available through flexible credit-based options.

Article Link: OpenAI Launches ChatGPT Agent That Can Complete Tasks For You
 
Article Link
https://www.macrumors.com/2025/07/17/openai-chatgpt-agent/
I like the idea of a Japanese breakfast for 4, but just as an idea. Instead, I have improvised a partially Mexican high protein breakfast for 1. It's unbelievably easy! You know what else was easy? Turning off autocorrect and autocomplete on my phone! Together, they were #$%^'ng up every sentence. I can do that myself!!🍸😹
 
  • Like
Reactions: Stenik
This is exciting development, however the performance is simply not good enough for productive work. By their own metrics on the public release update... it shows Agentic Spreadsheet and Agentic Browser both significantly weaker than human. I'm all for the agile approach - release something unpolished - better than release nothing... but personally I am not looking for a handicapped assistant.
 
  • Like
Reactions: Stenik
They should instead focus on making it accurate, even basic questions can produce the dumbest results. ChatGPT is useless, no point using an AI where you have to double check everything because the answer it gives isn't reliable. Now that the honeymoon phase is over, all I see with ChatGPT is a frustrating promise that is going nowhere.
 
  • Like
  • Disagree
Reactions: Alex.Roams, Stenik, tk421 and 1 other person
Facts are stubborn things. You can like or dislike Musk personally, but Grok 4's performance is totally awesome.

Grok 4 has is totally dominant on certain AI benchmarks -- as of today. Grok 4 Heavy scored a 51% rating on Humanity's Last Exam. All other AIs were scoring in the 20s on that test. If you're interested in the actual facts, you can review Matthew Berman's video about Grok 4 from last week. Matthew reviews ALL of the AIs.

 
  • Like
  • Disagree
  • Haha
Reactions: ducknalddon, Mackilroy, GubbyMan and 13 others
bobbyjack said:
They should instead focus on making it accurate, even basic questions can produce the dumbest results. ChatGPT is useless, no point using an AI where you have to double check everything because the answer it gives isn't reliable. Now that the honeymoon phase is over, all I see with ChatGPT is a frustrating promise that is going nowhere.
Click to expand...
Though that's no worse than asking a human. I have to double check everything I get from both a computer and a human. But at least the results given are a starting point for me.
 
Plan and buy ingredients to make Japanese breakfast for four.
Click to expand...
This is something I'd enjoy researching for myself. This to me is an example of how AI can produce lazy minds devoid of deeper inquiry. Why bother finding out things for yourself when you can just have the computer tell you what to do? I guess it's all well and good -- if you trust the computer.
 
  • Like
Reactions: ducknalddon and rnb2
Apple’s in SUCH a bad position as OpenAI’s new whirligig requires a device and, doggone it, Apple sells those!

Apple doesn’t have to rush for a solution because anything the competition makes, will run on their devices. :)
 
waziazi said:
amazing what trash siri and apple intelligence has become. the company is literally worthless at this point. apple has no purpose.
Click to expand...
I hate Apple too. But your post is de facto spam and factually incorrect - they're a multi-trillion dollar company. That's Trillion with a T.


macfacts said:
You didn't say Simon says, meaning you have to ask these AI agents in a very specific way to get the desired results.
Click to expand...
Die Hard With a Vengeance!!
 
  • Like
Reactions: racerhomie and tranceme
bobbyjack said:
They should instead focus on making it accurate, even basic questions can produce the dumbest results. ChatGPT is useless, no point using an AI where you have to double check everything because the answer it gives isn't reliable. Now that the honeymoon phase is over, all I see with ChatGPT is a frustrating promise that is going nowhere.
Click to expand...

It can’t ever be accurate or trustworthy. That’s not how it works. The entire industry is hedging on making it good enough but it’s not going to happen. Users tire of it quickly and distrust it as you have found.

A key sign of failure is when you see how many users they quote. What is more important is user retention and that’s not a good number or they’d be publishing it.

Regarding the first point about it not happening this is usually explained away with logarithmic scales on charts by analysts and dubious test systems by the producers. All backed up by “papers” which wouldn’t get a pass grade on an undergraduate dissertation.

As I’ve posted elsewhere the professional investors are out and stock is being moved to private bag holders and ETFs where the end investor pays the loss.

It’s going to go bang in a big way soon.
 
  • Like
  • Disagree
Reactions: 01cowherd, CarpalMac, freedomlinux and 5 others
You must log in or register to reply here.
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.
Top Bottom
Back