# OpenAI's Dot agent reviewed: capable with your own files, tripped up by the open web

A hands-on test of OpenAI's new Dots agent finds it strongest when given access to a user's own computer and files, but repeatedly blocked by security checks and lacking the checkout integrations of rival agents.

Canonical URL: https://freelancenews.online/news/openai-s-dot-agent-reviewed-capable-with-your-own-files-tripped-up-by-b19a7eb5
Published: 2026-10-02T19:17:48.131Z
Updated: 2026-10-02T19:17:48.131Z
Source published: 2026-10-02T18:00:00.000Z
Event date: Not established
Review status: source-reviewed
Review method: Automated comparison against retrieved source text; not independent fact-checking.

## Report

OpenAI's new agent platform, Dots, is being positioned less as a personal shopping assistant and more as workplace software that happens to be able to run errands, according to a hands-on review published by The Verge. The reviewer, who tested the product on OpenAI's $100-per-month Pro tier, described the experience as closer to "Codex, but for regular people" — a tool aimed at getting work done rather than a consumer companion.

According to the review, OpenAI announced Dots earlier in the week that the piece was written. The reviewer reports that Dots are being made available by OpenAI initially to those on its top-tier accounts, and that includes the Pro plan used for testing. Despite customizable names and blobby, anthropomorphic avatars, this gated rollout signals, the review argues, who the product is built for. A single Dot is what users currently get, although multiple Dots will be possible in the future, OpenAI says.

Mechanically, the interface resembles rival agent products: the user chats with the agent in one window while watching it click around a virtual machine in another. Out of the box, a Dot's virtual machine can access applications including Blender and GIMP, and users can grant it access to their own computer through the desktop ChatGPT app. The reviewer also used a voice call feature to talk through tasks out loud.

The review's most concrete finding is a split between controlled and uncontrolled environments. When the reviewer gave the Dot access to a personal website, it produced a new, more streamlined design, and the reviewer iterated on it over roughly ten minutes of voice conversation, walking away to do dishes before being pinged that the next version was ready. The reviewer characterized the result as not perfect but better, and noted that the agent proved most helpful when fed a large amount of unstructured input it could turn into something usable.

The same pattern held when the Dot was given access to the reviewer's own computer. Two tasks were assigned: consolidating several video files on the desktop into a single clip sized for social media, and deploying the website changes worked through in the prototype. Both succeeded, though the reviewer had to grant what was described as a shocking number of permissions. The reviewer watched the agent navigate the website backend and rebuild the prototype block by block, then received phone updates while away from the computer.

Against that, the agent struggled on the open web. Asked to schedule an internet service installation appointment, it nearly succeeded and even surfaced a $100 promotional discount buried in the reviewer's email, but stalled at a human verification check. The agent reported that the check required a sustained mouse hold its browser controls did not support, and asked the reviewer to complete it manually. The reviewer complied, comparing the experience to teenagers asking an adult to buy them wine coolers.

The installation task then reached a checkout screen requiring bank account details. The review notes there is no Stripe-style integration comparable to what rival Meta Muse offers for saving card information and using a virtual card at checkout. The reviewer declined to enter checking account numbers into the virtual browser and completed the task personally instead.

Other personal tasks produced mixed results. The Dot failed to find a free tour and entry option at a coworking space and instead offered a $35 day pass; a competing agent, Instinct, found the trial option and booked an appointment within minutes. Asked why it missed the option, the Dot said it had jumped to the wrong conclusion. It also failed to log into an Ikea account after getting stuck in a looping security check, and a local restaurant's ordering page blocked it entirely.

The reviewer observed that something about the Dot's footprint appears to trip security checks more often than Muse or Instinct, and that it required more manual input and browser takeovers than other agents — possibly because it was not purpose-built for ordering food or handling retail returns. The Dot itself pointed to an OpenAI page about websites blocking its cloud browser traffic as an explanation.

The review frames the product's identity as a deliberate contrast with consumer agents. Where personal agents like Muse are characterized as glorified autofill and a universal shopping cart, the Dot is described as a harbinger of a future of work. Because users pay upfront, the review reasons, the product may avoid the monetization pressure that consumer agents will eventually face — a logic the reviewer says fits enterprise software, since nobody wants to be nagged by AI while working.

The reviewer's verdict is qualified. The Dot handled tasks that, by the reviewer's own admission, would otherwise have gone undone, and it did eventually order teriyaki after some help logging into an Uber Eats account. But the reviewer did not consider it worth $100 a month for their own line of work, while allowing it could be far more useful to someone in the thick of launching a business.

For freelancers, designers and developers, the review suggests the practical value of this class of agent currently concentrates where the work happens inside files and systems the user already controls — code, assets, site backends — rather than across third-party sites that actively resist automated browsers. The permission burden and the frequency of manual handoffs are the tradeoffs to weigh against the time saved. The reviewer's successful outcomes were not limited to one setting: the personal-website redesign and the desktop video and deployment tasks both worked, and both depended on granting the agent broad access to the reviewer's own environment.

Several questions remain unresolved by the evidence. The review covers a single user's experience on a paid tier during an early, gated rollout, so it does not establish how the agent performs across accounts, regions or task types. Pricing beyond the $100-per-month Pro tier is not addressed, the promised support for multiple Dots is described only as a future capability, and the security-check failures are attributed by the reviewer to the agent's footprint without a confirmed technical cause. The comparison with Muse and Instinct reflects one reviewer's testing, not a controlled benchmark.

The clearest supported conclusion is that OpenAI's Dot behaves like workplace software with a consumer-friendly face: effective when pointed at a user's own environment with sufficient permissions, and unreliable when left to navigate the open web alone. Whether that tradeoff is acceptable depends on the kind of work being delegated and how much manual intervention the user is willing to absorb.

## Key points

- OpenAI's Dots agent is being offered first to users on its highest-tier accounts, including the $100-per-month Pro plan used in the review.
- The agent succeeded at consolidating desktop video files and deploying website changes when given access to the reviewer's own computer, though it required extensive permissions.
- It repeatedly stalled on security checks and human verification prompts while attempting tasks on third-party sites, including an internet installation booking and an Ikea login.
- The review notes the absence of a Stripe-style checkout integration comparable to rival Meta Muse, forcing manual entry of payment details.
- The reviewer judged the agent useful for tasks that would otherwise go undone but not worth $100 a month for their own line of work.

## Practical implications — editorial interpretation

For freelancers and developers, the review indicates that agent tools of this type deliver the most reliable value when applied to assets and systems you already control — local files, code, site backends — rather than to third-party sites that block automated browsers. Budgeting for a $100-per-month tier should account for frequent manual handoffs and broad permission grants, which the reviewer found necessary for the successful tasks.

## Limitations and unknowns

This is one reviewer's hands-on account of an early, gated rollout on a single paid tier, not a controlled benchmark. Pricing beyond the Pro tier, performance across accounts and regions, and the promised multi-Dot capability are not established by the evidence. The security-check failures are attributed to the agent's footprint without a confirmed technical cause, and comparisons with Muse and Instinct reflect the same reviewer's testing.

## Sources

- [1] theverge.com: OpenAI’s Dot agent is enterprise software that can also order your dinner
  https://www.theverge.com/ai-artificial-intelligence/1004096/openai-chatgpt-dots-hands-on-agent
  Retrieved: 2026-10-02T19:17:23.565Z

## Claim references

- OpenAI is offering Dots first to users on its highest-tier accounts, including the $100-per-month Pro plan the reviewer used. [source 1]
- The Dot's virtual machine can access apps like Blender and GIMP, and users can grant it access to their own computer via the desktop ChatGPT app. [source 1]
- The agent stalled at a human verification check requiring a sustained mouse hold its browser controls did not support. [source 1]
- The review states there is no Stripe-style integration comparable to Muse's for saving card details and using a virtual card at checkout. [source 1]
- Given access to the reviewer's computer, the Dot consolidated video files and deployed website changes, though it required a large number of permissions. [source 1]
- The reviewer concluded the agent was not worth $100 a month for their line of work, while seeing how it could be more useful to someone launching a business. [source 1]