The short version
OpenAI’s Dots are live for $100-a-month ChatGPT Pro subscribers, and The Verge has now taken one for a real-world spin. The verdict: Dot works best as an enterprise coworker on projects it controls — and stumbles on the messy, verification-heavy open web. It can run your desktop apps, take a phone call, and yes, it can order your dinner.
October 3, 2026 — Tested Intelligence
What happened
OpenAI introduced Dots at its DevDay keynote on September 29: persistent AI agents with blobby, anthropomorphic avatars and customizable names, each running on its own cloud-hosted computer. You chat with your Dot in one window and watch it work in another as it clicks around a virtual machine preloaded with apps like Blender, GIMP, Inkscape, and a Chromium browser — and you can call it by voice when you want to talk through a problem out loud.
This week The Verge put one through real-world tasks — using an expensed ChatGPT Pro account, since Dots are gated behind OpenAI’s premium tiers. Right now every user gets exactly one Dot, with multiple agents planned for the future. Access also extends to Business Premium and Enterprise plans in eligible markets.
Where Dot shines — and where it trips
The strong showing came on controlled projects. Asked to redesign a personal website, the Dot iterated through multiple rounds of revisions from spoken instructions, producing meaningful design updates each pass. Voice-plus-text refinement with full context is clearly where the platform is at its best.
Everyday errands were a different story. Asked to schedule an internet provider installation, the agent made it through most of the process but kept getting stuck on security verification challenges that required human intervention. It had trouble finding promotional offers on some sites, and its cloud browser footprint triggered anti-bot protections more often than competing agents — especially on e-commerce and service platforms.
The positioning tells you who OpenAI is really building for. Meta’s Muse and the viral agent Instinct are free; Dots cost $100 a month. Cute avatar aside, this is enterprise software that happens to be able to order you a burrito.
Why it matters
The Dot review is a useful reality check on the agent race. Giving an AI its own computer is genuinely new — an agent that can open Blender or push through a redesign by voice call is a different class of tool than a chatbot that can only text back. But the open web wasn’t built for agents: CAPTCHAs, fraud checks, and session guards treat a cloud VM like a bot because it behaves like one. OpenAI’s bet is that controlled, enterprise workflows are the profitable proving ground, and the free competitors will have to answer with reliability of their own. If you’re a subscriber wondering whether the $100 tier is worth it for this alone — right now, it’s a promising coworker, not an errand-runner.
Quick answers
What is OpenAI’s Dot?
A persistent AI agent announced at DevDay 2026 that runs on its own cloud computer, operates desktop software like Blender and GIMP, and can connect to your own computer through the ChatGPT desktop app.
How much does Dot cost?
Access starts with ChatGPT Pro at $100 per month, plus Business Premium and Enterprise plans. You get one Dot per account for now.
Is Dot better than Meta’s Muse or Instinct?
It depends on the job. The Verge’s testing found Dot stronger at structured projects under your control, and weaker at navigating third-party websites — plus Muse and Instinct are free.
Can Dot access my computer?
It runs primarily in its own cloud VM, but you can grant it access to your personal computer through the desktop ChatGPT app, and place voice calls with it.
Sources: The Verge (via syndicated summary); Wccftech on the Dot VM’s AMD EPYC-based compute and preloaded app list.