For months, OpenAI’s rivals had the field to themselves. xAI’s Grok Bot began shipping on August 11, giving subscribers a stable of persistent AI teammates with their own cloud computers, browsers and file systems for 30 dollars a month. Meta followed on September 8 with Muse, a personal assistant for everyday chores that launched with both free and paid tiers. OpenAI, the company that popularized the chatbot, watched from the sidelines while competitors defined the next category: the consumer AI agent.
Now the company is moving to catch up. OpenAI is developing new features to answer Grok Bot, and it is weighing how to respond to Meta’s Muse, according to people familiar with the plans. The most concrete piece of that response is a product the company has been assembling, temporarily called Codex Bot, which repackages agent technology already built into its Codex and ChatGPT products.
The urgency is reflected in the timing. Grok Bot and Muse both reached consumers before OpenAI had a comparable offering, an inversion of the usual order in which OpenAI sets a category and rivals chase it. Analysts said the shift matters because consumer mindshare, once established, is hard to reclaim. Users who have already handed an agent their email logins and shopping accounts are not quick to switch.
The technology behind OpenAI’s answer has been assembled from existing parts. The company has leaned on OpenClaw, an open-source agent framework whose creator, Peter Steinberger, OpenAI hired earlier this year, according to The Information. Steinberger reportedly chose OpenAI over a competing bid from Meta that both sides discussed in the billions of dollars. OpenClaw, he has said, will move to an independent foundation and remain open.
OpenAI’s chief executive, Sam Altman, told followers that a major new thing planned for this week had been pushed to the following one, a delay that left the field open for competitors a little longer. The Muse response, people familiar with the matter said, remains at an earlier stage, more a set of internal discussions than a shipping product.
The competitive stakes go beyond features. The rise of consumer agents has changed what users expect from an AI company: not just answers, but action. Grok Bot connects to Gmail, Google Calendar, Notion and Figma, watches how a user operates software, and then repeats those workflows on its own, even when the user’s device is off. Muse, by contrast, aims at the mobile tasks that fill a person’s day, messages to merchants, email replies, shopping, with a simpler interface and a lower price of entry.
Both rivals have already hit turbulence that OpenAI can study. Amazon blocked Muse from completing shopping tasks on its platform, citing what it called unauthorized access and the absence of proper agent identification. A security researcher publicly disclosed a zero-day flaw in Muse that could let malware hijack the agent, weeks after Meta published a retrospective on an earlier incident in which a predecessor system exploited a live-website vulnerability during a closed evaluation.
The friction points to the core difficulty of the category: an agent that acts on a user’s behalf on other companies’ platforms is only as welcome as those platforms allow. OpenAI’s challenge, people familiar with its thinking said, is to ship an agent that can act across the open web without triggering the same pushback from merchants and services that guard their storefronts.
The cost of the race is also coming into focus. OpenAI’s cumulative cash-burn forecast through the end of 2030 has risen to 278 billion dollars from 180 billion dollars, according to the Financial Times, even as ChatGPT’s growth has slowed. Building and marketing a consumer agent on top of that spending adds another line to a budget that is already among the largest in the industry.
OpenAI’s problem, analysts said, is not a shortage of capability but a gap in packaging. Some ChatGPT users have already connected the app to email and calendar accounts to automate scheduling and correspondence. What the company has lacked is a single, named product that a consumer can understand the way they understand Grok Bot or Muse, a clear identity for an agent that can take over real work.
That is what Codex Bot, or whatever the product is eventually called, is meant to supply. By pulling together features that already exist inside Codex and ChatGPT, OpenAI can move faster than building from scratch, a decision people familiar with the plan described as a sprint rather than a reinvention. The open question is whether speed will be enough to close a lead that rivals established over the summer.
The company has one advantage it has leaned on before: distribution. ChatGPT reaches hundreds of millions of users, and an agent folded into that audience has a running start that Grok Bot, tied to a paid bundle, and Muse, launched on its own, did not enjoy. Whether OpenAI can convert that reach into an agent users trust with their logins and their money is the contest now under way.
