Even Altman thinks the name ChatGPT Work is too small?
ChatGPT Work is amazing, and the name "Work" undersells it.
Late on July 26, Altman said this on X.
Before posting, he had just used his phone to ask ChatGPT to arrange a long weekend trip for himself and 8 friends.
His instruction was:
Use all my chat history to come up with ideas for a long weekend with me and 8 friends, plan the best three options, build a full-stack website where the 9 of us can discuss what to do at each place and decide where to go, and then book once we agree. Also, draft an email in my Gmail that I can send to friends once the site is ready.
Altman said it just worked.
Tibo Sottiaux, head of Codex and Work, retweeted this.
He said ChatGPT now handles about twenty tasks for him every day: negotiating internet fees with carriers, unsubscribing from spam emails, and searching for product deals.
None of them are work-related, but Work handles them all.
That's why a few days ago, Tibo joked on X whether ChatGPT Work should be renamed "ChatGPT HelpMeWithEverything."
OpenAI President Greg Brockman retweeted Tibo's post, saying ChatGPT is increasingly becoming your personal AGI.
Less than a month after the product launch, three executives have spoken out in succession.
They are all saying the same thing: the name Work is too narrow; ChatGPT Work is targeting more than just your job.
Say one sentence to your phone, and it can build a 9-person travel website
If you break down Altman's instruction, ChatGPT did 5 completely different tasks for him.
"Use all chat history to find ideas": The starting point was Altman's entire chat history—what he likes, where he's been, who he hangs out with.
"Plan the best three options": Based on that chat history, have ChatGPT generate three options to choose from.
"Build a full-stack website": For the 9 people to coordinate what they want to do and decide where to go.
"Book": Once everyone agrees, make the booking.
"Draft an email": Draft an email in Altman's Gmail, so he can send it to friends once the site is ready.
From retrieving memories, generating plans, building a site, waiting for consensus, to finally producing a sendable email—just say one sentence to your phone, and ChatGPT can do all that for him.
These 5 tasks span 3 types of work: retrieving from memory, building an interactive website from scratch, and writing a draft email in his mailbox.
From giving you answers to completing tasks
A year ago, if you asked an AI to plan a trip for 9 people, it would at most give you a guide.
Now, it can build a website that all 9 people can access, express opinions, and modify plans.
The feature supporting this action is called Sites, which entered public beta with ChatGPT Work.
Officially, it is positioned to turn results or ideas into interactive sites and web apps, shareable via a URL with a team or everyone, suitable for real-time data dashboards, project timelines, release schedules, prototypes, internal portals, and interactive reports.
More importantly, once the page is built, it doesn't stay static. If the data it references changes, ChatGPT can automatically go back and update the content.
The Sites example shown officially, an interactive revenue forecast dashboard that automatically updates when underlying data changes.
The barriers in front of ordinary people—writing code, deploying, and configuring domain names—have now been leveled by AI.
What makes it work are these capabilities of ChatGPT Work
On July 9, OpenAI launched the agent ChatGPT Work, targeting tasks that are time-consuming, multi-step, and require tangible outputs.
For complex projects, it breaks them down into smaller steps and can work continuously for hours if needed.
ChatGPT desktop menu, with Work listed alongside Codex, Scheduled, and Sites.
Its underlying technology is Codex, driven by the concurrently released GPT-5.6. Also updated are the following capabilities:
Plugins connect ChatGPT to external systems like Gmail, Calendar, Google Drive, and Slack. You can also use @ mentions in prompts to fetch data from specific apps;
Sites handles website building, providing real-time dashboards, project trackers, and interactive reports;
Accompanying features include Scheduled Tasks and automatic review: Scheduled Tasks run repeatedly on a schedule or based on triggers; automatic review checks before key actions to prevent sensitive information from being sent out inadvertently.
ChatGPT plugin directory, already integrated with Google Drive, Gmail, Teams, Slack, Salesforce, and other tools.
Codex is now used by over 5 million people weekly, with over 1 million using it for tasks other than coding: this product was originally a programming agent for developers.
OpenAI thus concluded that its capabilities are fully sufficient for a broader range of tasks.
OpenAI's internal adoption rate is even more telling: nearly 100% of teams, including finance and sales, use Work and Codex.
Dump the most annoying chores on ChatGPT Work, and it handles them all
After ChatGPT Work launched, TechRadar's Graham Barlow threw all the household chores he least wanted to touch at it.
After a week, the results exceeded his expectations.
For example, asking it to book a restaurant for a wedding anniversary, it returned three decent options—and that was the easiest task.
For vacation planning, Graham fed in dates, budget, and family requirements. It researched destinations, compared flights and accommodations, created a cost table, generated an itinerary, and even made a checklist indicating what was booked and what wasn't, updating as progress was made. It also checked existing bookings in Gmail.
What Graham appreciated most was the transparency of the entire process: what it was doing, which sources it referenced—every step was viewable.
For home renovation, it compared quotes, reviewed drawings and product specs, set timelines, and tracked pending decisions.
Graham found the most useful feature to be scheduling the family's next week's agenda.
This was made possible by the aforementioned Scheduled Tasks feature.
The scheduled tasks interface on mobile, allowing repeated execution on a schedule or continuous monitoring of changes.
ChatGPT used to have an old problem: it couldn't execute tasks on a schedule by itself; if you didn't ask, it wouldn't act. Now it can run repeatedly on a schedule and also monitor changes for you.
After connecting calendar and email, Graham set a scheduled task to automatically generate the next week's family plan every Sunday evening, with alerts for any changes. For example, school activities are most easily missed, but as long as the school sends an email, it won't be overlooked.
After trying it out, Graham felt that even an ordinary personal task could be complex enough to warrant using an agent.
He also saw a new use for ChatGPT Work: having it oversee a task from start to finish, regardless of whether it's work-related.
The phone becomes the remote control for the agent
Throughout the entire chain, Altman only did three things: describe the goal, review the plan, and approve key operations.
All three are decisions; execution was entirely left to ChatGPT.
The official demo shows ChatGPT Work's desktop and mobile versions, which have been integrated with Google Calendar, Slack, and Gmail.
So, on the execution side, where is the boundary of AI?
The first boundary is what it can read.
Altman wrote in the instruction, "Use all my chat history," but that's just his request, not necessarily what the system can actually do.
The official description of Memory has always been to synthesize useful context from chats, files, and connected applications, without promising to read every single piece of history, and users can turn it off at any time.
As for what Work can access, when it asks you, and whether it needs your approval before taking action, it's determined by you and your workspace admin. It doesn't have default access to all your accounts.
The second boundary is what it can do.
Altman asked ChatGPT Work to draft an email for him. He wanted a draft, which he would review and send himself.
OpenAI's sample prompt for email tasks on the Work product page is exactly like that: find emails that need replies, summarize each, draft replies, keep them as unsent, and wait for review. The official release also states that key operations involving connected tools require approval before execution.
Another detail worth noting: Altman's instruction was sent from his phone.
The role of the phone has changed; it is becoming the remote control for the agent.
Work on the web and mobile runs in the cloud, with sessions synced across three endpoints: you start a task on the go, review drafts on the go, check progress during a meeting break, and when you return to your desk, you can continue the same task on the web.
Barlow's assessment: ChatGPT is starting to become a place where you can advance a project at any time.
And that project might not be work-related.
Brockman said that ChatGPT is increasingly becoming your personal AGI.
For this vision to truly materialize, the hardest step may not be whether the model is smart enough or whether there's enough computing power.
Authorization, payment, personal data... these things that require human approval have not been eliminated yet.
References:
https://x.com/sama/status/2081396796174282900
https://x.com/gdb/status/2081458174662726009
https://x.com/thsottiaux/status/2081444811647963244
https://x.com/thsottiaux/status/2079731272797372425
https://t.co/zUQfWra6in
This article is from WeChat public account "Xin Zhi Yuan," author: ASI Revelation
