OpenAI Launches Dots: High-Tier AI Agents Tailored for Workplace Tasks
OpenAI has introduced Dots, a premium agent platform designed for enterprise tasks. While highly capable at software workflows, early tests show it still struggles with web security checks.
Following the recent wave of AI agent announcements, OpenAI has introduced "Dots," a new agent platform designed to perform tasks on a virtual machine. While these agents sport customizable names and cartoonish, blob-like avatars, they behave less like casual personal shoppers and more like enterprise-focused digital coworkers. Currently, OpenAI is rolling out Dots to its highest-tier subscribers, starting with those on the $100-per-month Pro plan. This premium pricing structure sets it apart from free, ad-supported alternatives like Meta Muse or Instinct.
Users interact with their Dot through a split-screen interface, chatting in one window while monitoring the agent’s actions on a virtual machine in another. Right out of the box, these agents can run applications like Blender and GIMP. They can also connect directly to a user's local computer via the desktop ChatGPT app or engage in real-time voice conversations to talk through complex tasks. However, early testing reveals that while Dots show promise with intensive, software-based tasks, they still struggle with everyday web-based errands.
Challenges with real-world automation
When tasked with simple consumer errands, OpenAI's agent frequently ran into security barriers and technical limitations. For instance, in an attempt to schedule an internet installation, the bot successfully found a deleted $100 promotional discount in the user's email but stalled at a security check. The agent explained, "This particular check needs a sustained mouse hold that my browser controls don’t support," requiring manual human intervention to proceed. Furthermore, the lack of integrated payment systems like Stripe meant the user had to manually enter sensitive banking details.
The agent also struggled with accuracy and web navigation. When asked to find a free trial for a coworking space, it missed the option entirely and recommended a $35 day pass, prompting the agent to later admit, "I jumped to the wrong conclusion." Meanwhile, security protocols on sites like Ikea and a local restaurant blocked the agent's cloud browser traffic entirely. OpenAI acknowledges these issues, pointing to documentation on how external websites block its cloud-based browser traffic.
Where the agent succeeds
Despite these setbacks, the platform proved highly capable when operating within controlled, software-heavy environments. When tasked with updating an outdated personal website, the agent successfully processed ten minutes of spoken feedback from a voice call to draft a prototype.
Once granted permission to access a local desktop, the agent also demonstrated its utility by compiling multiple video files into a single, correctly formatted clip for social media and deploying website updates block-by-block. In these scenarios, the agent acts more like an accessible version of Codex, handling tedious technical workflows in the background while the user focuses on other tasks.
What it means for developers
The launch of Dots highlights a significant shift in how AI companies view the future of productivity. Rather than relying solely on chat interfaces, the industry is moving toward agentic workflows that can control virtual environments, run desktop applications, and execute multi-step tasks.
For developers, this evolution brings both opportunities and challenges. Building agents that can reliably navigate the web requires overcoming persistent obstacles like CAPTCHAs, security blocks, and the lack of standardized payment integrations. Developers must also design systems that gracefully transition control back to human users when automated steps fail.
As the landscape of agentic tools expands, choosing the right underlying model for specific tasks becomes crucial. Developers looking to experiment with different agentic capabilities can try top AI models cheaply through one API at https://apixoai.online, simplifying the integration of diverse models into their own applications. Ultimately, OpenAI's Dots demonstrate that while fully autonomous personal assistants are still hindered by web security, the era of the digital coworker is rapidly approaching.
Source: OpenAI’s Dot agent is enterprise software that can also order your dinner — The Verge AI. Written by the Apixo team from that report.
One key for Claude, GPT, GLM, DeepSeek and more. Pay per token with crypto.
Get your API key
