OpenAI launches Dots, an enterprise AI agent that runs apps, browses and takes calls

OpenAI has begun rolling out Dots, a new AI agent platform that the company is positioning as workplace software rather than a personal shopping helper. According to a hands-on review published by The Verge on October 2, the product pairs a cartoonish avatar with a cloud-based computer that can operate other software on the user's behalf.

What OpenAI shipped

Per The Verge, OpenAI announced Dots this week. Each user currently gets a single agent — support for several is planned — with a customizable name and a blob-like anthropomorphic look reminiscent of Meta's Muse. The interface puts a chat window alongside a live view of the agent working in a virtual machine, where it can use applications such as Blender and GIMP from the start. OpenAI claims the agent can handle almost any task given its cloud computer and app access, The Verge reports.

Two capabilities distinguish it from a chatbot. Users can grant a Dot access to their own machine through the desktop ChatGPT app, and they can place a phone call to the agent to think through a task out loud.

Access is limited for now to top-tier subscribers, including the $100-per-month Pro plan. That contrasts with Muse and the rival agent Instinct, which are currently free — a gating choice The Verge reads as a clear statement about the intended audience: professionals rather than casual users.

Everyday errands expose the limits

The Verge's reviewer, Allison Johnson, put Dot through mundane personal tasks with uneven results. Asked to schedule an internet installation, the agent progressed far enough to surface a $100 promotional discount buried in her email, but stalled at a human-verification button that required a sustained mouse hold its browser controls could not perform. She had to take over there, and again at checkout when the provider wanted bank account details, which she chose to enter herself rather than hand to the agent.

It also overlooked a free tour option at a coworking space and proposed a $35 day pass instead — a task the competing Instinct agent completed within minutes, according to the review. An attempt to log into her Ikea account got caught in a looping security check, and a restaurant blocked the agent from its ordering page entirely. The Verge observes that Dot trips security measures more often than Muse or Instinct, and notes that OpenAI maintains a page acknowledging that websites block its cloud browser traffic.

Where the agent earns its keep

Dot fared best on work the reviewer fully controlled. Asked to modernize her outdated personal website, it produced a streamlined redesign, then iterated on it during a 10-minute voice call while she washed dishes. Given access to her computer — after what she describes as a startling number of permission prompts — it deployed those changes to the live site and stitched several desktop video files into one clip sized for social media, sending progress updates to her phone while she stepped away.

The Verge frames Dots as essentially a mass-market counterpart to Codex: a delegate that takes a chunk of work off your plate while you do something else. Because subscribers pay upfront, the review argues, the product should face less pressure toward the ad-driven monetization its free rivals will likely eventually pursue. Even so, Johnson doubts it justifies $100 a month for her own line of work, while suggesting it could pay off for someone in the thick of running a business. And with a little help logging into Uber Eats, the agent did eventually place her teriyaki order.

Why it matters

Dots is OpenAI's clearest signal yet that it sees agents operating real software — not just answering questions — as its next product category, and the Pro-first rollout shows the company betting on professional subscriptions as the business model. The review's failure cases also point to a structural problem for every agent: anti-bot defenses, payment flows and login walls are precisely where autonomous software breaks down and hands work back to humans. Whether OpenAI can negotiate that friction, or whether websites adapt to accommodate agents at all, will go a long way toward deciding whether Dots becomes a genuine coworker or a novelty.