OpenAI adds business-run agents to ChatGPT
Companies can track mentions and clicks, but OpenAI says it won’t share your chats with them.
By Saeed Ezzati8 min read
The audio edition
Listen to this newsletter
0:005:14
Read transcript
OpenAI is giving businesses a verified place in ChatGPT—and, for some of them, a way to answer customers directly. Its new Businesses in ChatGPT beta starts with Business Hub, where eligible companies can verify their organization, update profile details and provide product feeds with current pricing and availability. The access rules differ by feature. Eligible businesses can claim profiles wherever ChatGPT is available. But building a customer-facing agent is initially limited to eligible ecommerce and online-service businesses in the United States. Shopify merchants can connect their stores to streamline verification; OpenAI says this does not introduce their product catalogs to ChatGPT, since eligible products were already integrated through Shopify Catalog. The customer’s handoff is optional. If a business appears in a ChatGPT response, a user can open its profile and choose to start a conversation with its agent, if one is available. OpenAI says those chats are clearly labeled and separate from the original ChatGPT conversation. Businesses can also sponsor their agents as destinations for ChatGPT Ads, with sponsored placements labeled and kept apart from ChatGPT’s answers. The business-facing dashboard is designed to show patterns, not private exchanges. OpenAI says companies receive aggregated mentions, website clicks and recurring discovery topics—not individual conversations or agent transcripts. Insights appear only after interaction thresholds are met, though the company hasn’t said what those thresholds are. OpenAI plans to add more markets and business categories, without giving a timetable. The important test will be whether businesses can keep the information agents rely on accurate, while customers can tell clearly when they’re talking to a company rather than ChatGPT. That question of who controls an AI-mediated interaction also sits behind Anthropic’s proposed IPO, which could put its public-market valuation above a trillion dollars. The company set a proposed price range of seventy to seventy-eight dollars a share, but there’s no confirmed trading date—and the share counts reported by two outlets don’t match. Yahoo Finance reports 825 million shares for sale; Axios reports about 801.4 million. At the top of the range, those figures imply roughly $64.4 billion versus $62.5 billion in proceeds. Neither account explains the discrepancy. They’re also distinct from the company’s market value: Axios estimates about $1.03 trillion at the top price if underwriters exercise their option, based on roughly 13.09 billion shares across stock classes. Include options and other potential shares, and the estimates diverge further: Axios puts the fully diluted value near $1.37 trillion, while Yahoo Finance gives a range up to $1.45 trillion. Yahoo Finance also reports a separate $26 billion private purchase of non-voting shares at the IPO price. Those buyers reportedly avoid the usual six-month lockup. So the headline valuation is still a proposal, not a completed public sale; watch for the final share count, pricing and offering terms before treating the trillion-dollar figure as settled. The same distinction between access and actual capability matters in Meta’s smartwatch update. Its Muse assistant is now available on Apple Watch and Wear OS, through text messages, dictation or voice calls. Apple Watch users also get a watch-face shortcut through an update to the Muse iOS app. This expands where people can reach Muse; it doesn’t announce new task capabilities. Muse is already designed to help with tasks such as managing email and meetings, booking reservations and ordering groceries. Those are existing assistant features, not proof that a task can now be completed entirely from a watch. The release establishes that users can contact Muse from their wrist, not that every step—from request to purchase or booking—happens there. And as AI moves from a chat window into the physical world, Ring’s announced security robots make the control and privacy boundaries more consequential. The company says its $250 indoor Drone Cam will patrol homes, while a $1,000 Rover Cam is aimed at workplaces. Neither has launched: Rover is scheduled for December sixteenth, 2026, and the drone for January twenty-seventh, 2027. The home drone sends video to Ring’s servers so its cloud-based AI Pilot can direct flight. Owners can set routes and schedules, or have it respond to sensor alerts, but they can’t steer it manually. Ring says it can land if its internet connection drops. It flies for about five to eight minutes per charge and requires Wi-Fi. That cloud design also has a privacy tradeoff. Ring says video is encrypted in transit and its server-held key is deleted after use, but the company temporarily has access to the key to run AI features. That is not end-to-end encryption. The Rover, designed for warehouses and factories, is scheduled to patrol for about thirty minutes before docking. Both products remain announcements, not available devices. Across these stories, the practical question is not just where AI appears, but what it can act on, who can see the interaction and what limits apply. Watch for those details in the products’ actual rollouts: they’ll show whether the new access comes with clear, usable control.
