The AI assistant sector is shifting rapidly from simple "chat and answer" functions to systems that "get things done for you," sparking a critical contest for control over user workflow entry points. SpaceXAI's Grok Bot has been live for over a month, Meta's Muse has climbed to the top of the App Store, and OpenAI—now tagged as the "follower"—is rushing to respond. It is reportedly preparing to launch a personal agent product codenamed "Aeon," potentially within days, to counter the pressure from both rivals.

According to a Bloomberg report from September 22, Grok Bot reached 418,000 weekly active users as of September 14, a 24% increase from the prior week. Meanwhile, Meta's Muse surged to the No. 2 spot on the U.S. App Store's free chart within two days of launch and had claimed the top position by September 22, pushing ChatGPT down the rankings.

In parallel, The Information reported on September 21 that OpenAI is developing features to more directly challenge Grok Bot, which SpaceXAI introduced in August. Internally, there are also discussions on how to respond to Meta's personal AI assistant Muse, which debuted this month. Sources familiar with the matter indicate that OpenAI plans to repackage and rebrand existing "agent" technology already present in Codex and ChatGPT, rather than building a new product from scratch.

AI engineer @dotey posted on social platform X that a product originally slated for release last week was delayed, with confirmation from Sam Altman, and the goal is now to unveil it before the San Francisco DevDay developer conference on September 29.

Industry analysts suggest that whoever secures the "workflow entry point" for users first will directly determine future user retention and monetization potential.

The Grok Bot: 418,000 Weekly Users, with Musk Calling Its Growth "Unprecedented"

Grok Bot was officially launched by SpaceXAI on August 11, positioned as an "always-on AI team." Its key differentiator from conventional AI assistants is that each bot operates with its own persistent cloud computer, storing files, browser sessions, and login credentials—meaning the bot continues running in the background even when the user powers down for the night.

Functionally, Grok Bot can automatically log into a user's Gmail, Google Calendar, Notion, Figma, financial software, and even legacy websites without APIs. It operates like a human, controlling the mouse, filling out forms, and executing multi-step tasks, only prompting the user when authorization is required. Multi-bot collaboration is its core selling point: users can run multiple bots simultaneously, each dedicated to tasks like inbox management or sales outreach, with a single "chief of staff" bot orchestrating the workflow and allowing bots to assign tasks among themselves within a group chat.

Grok Bot is available through a SuperGrok subscription starting at $30 per month, along with a standalone app version. SpaceXAI positions it as an "AI digital chief assistant" for white-collar workers and small business owners, with aggressive advertising across social media, YouTube, and podcast platforms.

Musk himself stated on platform X: "The usage growth of Grok Bot exceeds anything we have ever seen."

Meta Muse: Topping the App Store with a Consumer-Focused Push

Meta launched its personal AI assistant Muse on September 8, targeting consumer scenarios. Its features cover daily tasks such as responding to emails on the user's behalf, communicating with merchants, price comparison for online shopping, and schedule management. Users can access Muse directly through Meta-owned apps like Messenger. Muse is free to download, creating a clear contrast with the enterprise subscription model of Grok Bot.

According to data from Bloomberg Intelligence, Muse's download numbers are four times higher than those of the standalone Grok Bot app. Meta holds a significant distribution advantage thanks to its app ecosystem's billions of users, and it has already given Muse prominent placement within core applications like Messenger.

However, a distribution edge does not guarantee success. The Information notes that Meta launched a ChatGPT-rival chatbot in 2024, but its massive user base failed to translate into the expected competitive advantage.

Meanwhile, the rise of invite-only startup Instinct is also drawing market attention. The product helps users book restaurant tables and shop via text messages and phone calls, requiring authorization to access credit card information and account passwords. Instinct currently boasts over 100,000 users and is seeking to raise funds at a $10 billion valuation—a figure that has quadrupled within a month.

OpenAI's "Aeon" Preparation: The Technology Exists, but Presentation Lags Behind

OpenAI is not without preparation, yet the market's awareness of its existing capabilities remains severely inadequate.

The Information reports that the product under development at OpenAI is internally codenamed "Codex Bot," built on the open-source agent framework OpenClaw. Notably, OpenClaw's founder, Peter Steinberger, joined OpenAI in February, and Sam Altman has explicitly stated that Steinberger is responsible for building the next generation of personal agents.

At the code level, leaked information has already hinted at the product's form. Tibor Blaho, a blogger who consistently tracks ChatGPT's code, discovered an "aeonId" field in two consecutive versions of the ChatGPT Android client, appearing alongside "accountUserId," with a neighboring field named "memberAeonIds." The use of "member" strongly aligns with Grok Bot's multi-bot group chat structure.

Test descriptions also include a reference to "aeon-style behavior," where capability status must be recaptured at each sampling step rather than waiting for the next round of user input. This perfectly matches the logic of an agent executing continuous multi-step tasks, suggesting that Aeon is not a simple feature iteration but a distinctly identifiable "digital member" that operates persistently.

In fact, OpenAI launched ChatGPT Work as early as July 9. Its official introduction already includes cross-app task execution, continuous processing of complex projects, and scheduled tasks, all integrated with Codex technology for cloud-based background operation. The issue is that many users are completely unaware these features exist—as The Information puts it: the kitchen can prepare any dish, but the menu only lists "chat."

According to @dotey's posts on X, Aeon is rumored to launch before DevDay, possibly even this week. OpenAI has not yet made any official comment on the matter.

The Battle for the Entry Point: Whoever Owns the Workflow Holds the Users

The essence of this competition is no longer a leaderboard race for model capability; it is a fight for control over users' daily workflow entry points.

According to The Information, four players are pursuing four distinct paths: SpaceXAI wants a cloud computer holding all the user's login states; Anthropic moved earlier on June 23 with Claude Tag, allowing Claude to join Slack as a "member" to receive tasks, targeting team-wide collaboration channels; Meta's Muse wants to be in the chat app users open every day; and Instinct simply asks users to hand over credit cards and passwords in exchange for the convenience of full delegation.

Once an entry point becomes a habit, the switching costs become extremely high. A bot that has taken over a user's inbox, learned their writing style, and remembers their priority clients is far harder to replace than a chat app.

For OpenAI, the pressure comes from multiple fronts. According to The Information, ChatGPT's growth has slowed noticeably over the past year. The target of 1 billion weekly active users was originally set for last year but was only achieved this summer. On the personal agent track—seen as the "second growth curve"—OpenAI has already fallen behind the market momentum of Grok Bot and Muse.

However, beyond technology and product design, the biggest uncertainty facing the entire sector may be user trust. How many everyday users are willing to grant AI access to personal information, social accounts, and even payment permissions remains an open question. In the chat era, users evaluated whether "the AI is smart enough"; in the agent era, users must answer, "Do I dare hand over the keys?"

This trust barrier is the final hurdle every participant must clear, and it will also be the decisive factor shaping the ultimate landscape of this entry-point war.