ChatGPT version of Grok Bot code revealed! OpenAI also aims to create AI colleagues who never leave work.

Your ChatGPT might soon have a new "person"!

Musk's bot has been running for over a month, Zuckerberg's Muse has topped the App Store charts, and an invitation-only small company is talking about a $10 billion valuation.

In just a few weeks, OpenAI has become a follower in the race of "AI doing work for humans".

According to The Information, OpenAI is developing features that are more directly comparable to Grok Bot, and is also internally discussing how to deal with Meta's personal assistant, Muse.

This report labels OpenAI as a "follower." A year and a half ago, this label would never have been applied to it.

When it launches Operator in January 2025, it will be the first company to turn "AI operating the browser for humans" into a consumer product.

ChatGPT has brought generative AI to ordinary people's phones, and according to The Information, its weekly active users exceeded 1 billion this summer.

After the report was published, Tibor Blaho, a blogger who has been digging into ChatGPT code for years, retweeted a post by journalist Stephanie Palazzolo and threw out one word: Aeon.

Aeon means "eternity" or "endless time" in Greek.

If this is indeed the name OpenAI has chosen for the intelligent agent, it's hard not to think that it's also going to create an AI colleague for you that never leaves work and is always online.

Tibor also attached a screenshot to the post, which was code he found in the ChatGPT client.

In the code of the ChatGPT web version, there is this line:

A message is recorded under the user's account if it was sent by the user; otherwise, it is recorded under the name Aeon.

In other words, in ChatGPT's view, Aeon and users are two parallel identities.

Just as it was being labeled a "chaser," leaked code revealed that it may have already reserved a workspace in ChatGPT for that AI colleague who can do the work for you.

Musk's bot has been working for over a month.

Why is OpenAI so restless?

Because Musk's bot has been working for over a month.

On August 11, SpaceX AI released Grok Bot. Its official description is minimalist: Your always-on intelligent agent team.

Note that you are being given a group of intelligent agents, not just one. The biggest difference between them and ordinary AI assistants is that they have their own computers.

According to the official description, all bots under the same username share a persistent computer hosted in the cloud, where files, browsers, and login status are all stored. Even when you close your laptop and go to sleep, it's still running.

These bots will automatically log into your Gmail, Notion, accounting software, and even old websites without APIs.

They operate the mouse and fill out forms like humans, doing all the work themselves. They only come to ask you a question when they need your approval.

Even more ruthless is the collaboration of multiple bots.

You can run several bots simultaneously: one to manage the inbox, another for sales and outreach, and a third as a "chief of staff" to coordinate everything. They can then assign tasks to each other in a group chat.

What Grok Bot is really selling is just this one action: handing over the work.

An extra "member" has appeared in the code.

In the code screenshot forwarded by Tibor, aeonId and accountUserId are listed side by side, indicating that in ChatGPT's system, Aeon may be some kind of object that can be identified individually, rather than a simple function switch.

Whether it has the same identity or permissions as the user is not visible in the screenshot.

In the two most recent consecutive versions on Android, aeonId appears repeatedly, along with a field called memberAeonIds.

"Member" is a name that inevitably evokes the image of multiple bots sharing a group chat, similar to Grok Bot.

Above the screenshot of Tibor, there is another test description:

This test scenario is specifically designed to protect against "aeon-like behavior," where capability states are recaptured at every sampling step, rather than waiting for the next round of user input.

In a typical ChatGPT, it's a back-and-forth exchange, while "aeon-style behavior" requires the system to reconfirm which capabilities are available at each step during multi-step execution.

This aligns with the logic required by an intelligent agent that needs to perform many steps continuously.

OpenAI has not yet made an official statement. All we have at the moment is a media report, a suspected code name, and a few lines of code, but the clues are already faintly visible.

ChatGPT's menu only lists "Chat".

Most intriguingly, media reports claim that OpenAI will not create a new product from scratch, but rather "repackage" the existing agent technologies from Codex and ChatGPT.

This statement reveals an embarrassing fact: the capabilities that competitors are now selling so well, OpenAI already had long ago.

On July 9th of this year, OpenAI released ChatGPT Work, which includes features such as cross-application task execution, continuous processing of complex projects, and scheduled tasks in its official introduction. It also integrates Codex technology.

The official documentation also states that if you choose to run the app in the cloud, the task will continue to run even after you close the app or close your computer.

Talent has also been planned in advance. In February of this year, Peter Steinberger, the author of OpenClaw, joined OpenAI. Altman said he would be responsible for building the next generation of personal intelligent agents.

The problem is that many OpenAI users may not be aware of these features.

This is like a kitchen that can cook anything, but the menu only lists "chatting".

If the rumors about Aeon are true, the real question worth asking is: on top of ChatGPT Work's existing execution capabilities, will it have an additional "digital member" that can be called by name and assigned tasks directly?

This is very likely its response to Grok Bot.

Why can't OpenAI afford to lose?

According to The Information, ChatGPT’s growth has slowed significantly over the past year, and OpenAI’s goal of 1 billion weekly active users, which it originally set last year, was only achieved this summer.

In this round, major AI companies are no longer vying for rankings on model leaderboards. Intelligent agents that can do the work for you have become the second growth curve that everyone is competing for.

Musk's Grok Bot is about creating a cloud computer that stores all your login statuses.

Anthropic had launched the Claude Tag earlier, on June 23, allowing Claude to join Slack as a member and take on gigs; they wanted the entire team's channel.

Meta's Muse climbed to number 2 on the US App Store's free app chart within two days of its release, and by September 22nd, it had reached number 1, pushing ChatGPT behind it. It's aiming to be the chat app you open every day.

Then there's Instinct, an invitation-only startup whose valuation quadrupled in a month. It helps you book restaurants and make purchases via text message and phone, requiring only your credit card and PIN.

Four competitors, four different strategies, all betting on your workflow.

Imagine a Monday morning to come: just after 8 a.m., you pick up your first cup of coffee of the day and glance at your phone.

In the past, you had to reply to a dozen emails, mute annoying group messages one by one, and finally cram a bunch of meetings into your calendar.

Later, by the time you finish this cup of coffee, all these things will probably be done. You'll only have a list in front of you waiting for your approval: whether to send this email, whether to pay this bill, whether to accept this meeting.

Whoever controls this list holds the key to your daily work.

Once you get used to something, it's hard to change it.

If a bot takes over your inbox, learns the tone of your emails, and remembers which customers need to be prioritized, getting rid of it is much harder than switching to a different chat app.

But the most difficult hurdle to overcome may lie outside of technology: trust.

How many ordinary people would actually allow an AI to access their personal information, social media accounts, or even authorize it to make payments on their behalf?

This hesitation is the final hurdle that all "AI colleagues" must overcome.

In the era of chat boxes, users asked: Is AI smart?

In the era of intelligent agents, users are asking: Do I dare to entrust the work to them?

There is only one question left for OpenAI: how to get 1 billion users to hand over their keys with confidence.

This article is sourced from: New Zhiyuan.

Risk warning and disclaimerInvesting involves risk; please exercise caution. This article does not constitute personal investment advice and does not take into account the specific investment objectives, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their specific circumstances. Any investment decisions made based on this information are at your own risk.