
OpenAI has officially launched a new AI agent platform called Dots, introducing customizable digital assistants designed to run virtual machines and perform tasks for users. Unlike competitors like Meta's Muse, which focus heavily on consumer activities like shopping, OpenAI’s platform is geared toward workplace and administrative applications. Currently, the tool is restricted to premium accounts, such as the company's one-hundred-dollar-per-month Pro tier, signaling a focus on business utility.
The system provides access to a single virtual assistant, represented by a customizable avatar. The interface is split into two windows: one for chatting with the assistant and another displaying the agent's actions as it navigates a cloud-based computer. These agents are pre-equipped with applications like Blender and GIMP. Users can also connect the agent to their local computers via the desktop ChatGPT app or communicate with it using voice calls.
Real-World Tests and Interface Hurdles
In everyday administrative tests, the agent delivered mixed results. When tasked with scheduling an internet installation, the agent successfully retrieved a one-hundred-dollar discount from deleted emails. However, its progress was halted by a security check requiring a sustained mouse hold, which its browser controls could not execute without manual user help. Lacking payment integrations like Stripe, the agent also required the user to manually enter banking details to complete the transaction.
The assistant struggled with other online navigation tasks as well. It failed to locate a free trial option at a coworking space, offering a thirty-five-dollar day pass instead. Similar issues arose during attempts to access an Ikea account, which triggered an infinite security loop, and while attempting to access a local restaurant menu, which blocked the agent's cloud traffic. OpenAI attributes these browsing failures to security protocols designed to block automated web traffic.
Technical Strengths and Media Editing
The agent proved highly capable when processing large amounts of information or managing local files. During a website redesign, the assistant processed a ten-minute voice call detailing design instructions and turned them into a prototype. After receiving system permissions, it accessed the website's backend to reconstruct the layout block by block.
It also excelled at local media editing. When asked to compile multiple video files from a desktop into a single clip sized for social media, the agent successfully formatted and outputted the files within minutes. It also managed to complete a restaurant order once granted manual assistance with logging into an Uber Eats account.
Why It Matters
The debut of Dots highlights a divergence in how AI firms monetize consumer agents. While rivals like Muse and Instinct are currently free, OpenAI is positioning its agents as paid utilities. This upfront pricing structure suggests that OpenAI intends to deliver an ad-free environment tailored for professional workflows, though at a premium cost that may limit its appeal to casual users.
Additionally, the agent's frequent struggles with web security reveal a major hurdle for the industry. As websites deploy stricter anti-bot protocols, autonomous agents will continue to hit roadblocks with transactional tasks. Until standardized authentication methods are established, these assistants will remain most reliable as internal productivity tools—handling coding, design, and media editing—rather than fully autonomous web navigators.
Source: The Verge
Original source: https://www.theverge.com/ai-artificial-intelligence/1004096/openai-chatgpt-dots-hands-on-agent