OpenAI's Dot agent is a business tool capable of ordering your dinner too.

OpenAI's Dot agent is a business tool capable of ordering your dinner too.
Summary
OpenAI's new agent platform, Dots, functions as a business-oriented AI assistant.
Dots can help with tasks but struggles with security checks and integrations.
The platform offers mixed results in personal tasks compared to competitors like Instinct.

Share

Bookmark

Newsletter

A senior reviewer with over ten years of experience in consumer technology journalism, particularly in mobile photography and telecommunications, has evaluated OpenAI's latest offering. OpenAI unveiled its new agent platform, dubbed Dots, which resembles charming digital assistants ready to follow your commands.

However, unlike the user-friendly Meta Muse, Dots operates more like office software that can also place lunch orders—putting a strong emphasis on productivity. This latest announcement from OpenAI reveals that Dots feature cute, customizable avatars and the capability to manage various tasks. Currently, you can deploy just one Dot, which I've affectionately named Dotty McDotface.

The Dots user interface is reminiscent of Muse, allowing you to chat with the agent in one window while it interacts with a virtual environment in another. However, instead of functioning primarily as a shopping assistant, Dots act more like colleagues efficiently navigating applications such as Blender and GIMP. Additionally, you can connect the agent to your own desktop via the ChatGPT app and even speak to it for collaborative problem-solving. Access to Dots is initially restricted to users subscribed to OpenAI’s higher-tier plans, such as the $100-a-month Pro account I utilized for this exploration. In contrast, Muse and its rival, Instinct, are currently free, although that may change once Meta fully launches its service and begins to monetize it.

I’ve been experimenting with various AI agents for everyday tasks and was eager to compare how OpenAI's bot measured up. Dots may exude a professional atmosphere, but OpenAI insists these agents are versatile, capable of handling nearly any task with their cloud computing and app accessibility.

I began my test by asking Dot to set up an installation appointment for a new internet service provider at my home. The agent made progress, even uncovering a $100 promotional offer hidden in my deleted emails, but it encountered a “human check" obstacle. “This check requires a sustained mouse click that my browser controls can’t manage,” it explained. It asked me to assist by holding the button down, which felt a bit odd; reminiscent of teenagers asking for help buying drinks.

With my manual intervention, Dot reached the checkout page but couldn’t facilitate my bank details securely like Muse’s streamlined Stripe integration. Instead, I opted to handle the bank information directly on my computer, which was a less than impressive start.

After assigning Dot several different personal tasks, outcomes varied. It struggled to locate a free trial for a coworking space I was interested in, mistakenly suggesting I sign up for a $35 day pass. I turned to Instinct’s agent for assistance, which promptly found the desired option. When I questioned Dot about its oversight, it admitted, “I jumped to the wrong conclusion.”

Next, I tasked Dot with retrieving information regarding an Ikea order and encountered a frustrating looping security check. Similarly, while trying to access a local teriyaki restaurant's ordering page, Dot was blocked due to policies that restrict its cloud browsing. It apparently has more trouble with security checks than Muse or Instinct, likely because it isn’t designed specifically for certain tasks, like placing food orders.

My experience turned favorable when I let Dot access my outdated personal website. I requested a simplified design, and after discussing my vision with it for roughly ten minutes, I was notified that a refined version was ready for review. While not flawless, it was a significant improvement, showcasing the effectiveness of providing a comprehensive dataset for the AI to work with. I continued to iterate on changes with Dot through both the browser and voice-to-text capabilities.

Once Dot had access to my computer, I assigned it two relatively straightforward tasks: creating a social media-sized video compilation from some footage and implementing the adjustments we discussed for my website. Although I had to grant numerous permissions, Dot executed the tasks successfully, displaying its ability to manage my website’s backend effectively. I received timely updates about the video and website adjustments, leading to a truly productive experience.

Reflecting on its capabilities, Dot serves better as a versatile assistant tailored for everyday business tasks rather than the more autonomous shopping helpers like Muse. The pricing model means it isn't likely to face the same advertising pressure, making it potentially less intrusive when fulfilling work-related needs. Still, its utility can vary greatly depending on specific job requirements and personal trust in OpenAI’s handling of tasks.

In the end, Dot proved helpful in accomplishing tasks I might have otherwise postponed, which is beneficial, though I’m hesitant to deem it worth the $100 monthly fee, given my current role. However, I can certainly see how someone launching a business could greatly benefit from its features. And with its final assistance, Dot successfully placed my teriyaki order through Uber Eats.

Loading comments...