It’s a story as old as last week: OpenAI’s new agent platform, called Dots, is full of cute little ones who can do your bidding.
But unlike the ultra-affordable Meta Muse, Dots is keen to use workplace software that can order you a burrito: emphasis on work.
OpenAI announced Dots earlier this week. Like Muse, Dots has anthropomorphic avatars and customizable names. In the future, OpenAI says you’ll be able to have multiple Dots, but now you have one. I called mine Dotty McDotface.
The interface is similar to Muse; You chat with the agent in one window and follow his work in another while he clicks on a virtual machine. But Dots are less personal shoppers, although they can help you buy things. They are more like co-workers: enterprise software that can use other software. Your Dot’s virtual machine can access applications like Blender and GIMP right out of the box. You can also give it access to your own computer via the ChatGPT desktop app, and you can call your Dot if you need to work things out out loud. And OpenAI is offering Dots first to users of its highest-tier accounts, including $100 ones.per month Pro account I spent to test it. Muse and its lively competitor Instinct cost you nothing, at least now, and at least until Meta can get it off the ground and presumably start using it to serve you ads, etc. But OpenAI’s closed launch says a lot about who the product is for, cute avatars notwithstanding.
I’ve been testing other AI agents with day-to-day personal tasks (with mixed results) and was curious how the OpenAI bot would compare. Despite its more entrepreneurial vibes, OpenAI says Dots “can do almost anything” with your cloud computing and app access, so I tried it with some more common use cases.
To start, I asked Dot to try to schedule an installation appointment for a new Internet service provider at my house (did you hear that, Comcast? You’re on notice). It almost got there, and even found a $100 promotional discount in my email that I had deleted en masse, but I got stuck on a “human check.” Which is problematic for a bot. “This particular check requires sustained mouse pressure which my browser controls do not support,” it said. “Open my browser here and hold ‘Push and Hold’ until it completes, then tell me and I’ll continue with the installation dates.” I agreed, but it felt a little like some teenagers outside the store asking me to buy them some wine coolers.
With my human ability to click and hold a button, Dot made it to the payment screen, where she asked me to enter my bank account details. It turns out there’s no fancy integration with Stripe, like Muse has for saving your credit card information and using a virtual card to pay. And in any case, the new supplier needed my bank account to get $5 off my bill each month. Dot paused to let me enter the numbers into the virtual browser; I didn’t trust it enough to enter my checking account number and decided to do it all myself on my own computer. Not exactly a spectacular start.
I gave Dot some other personal tasks and the results were mixed. He couldn’t find the option for a free tour and admission to a coworking space I wanted to try and instead offered to sign me up for a $35 day pass. I had the Instinct agent, which I’ve also been testing, look for the trial option; He found it and made an appointment within minutes. “Why did you lose this option?” I questioned Dot over the phone after the incident. “I came to the wrong conclusion,” he said. I hung up.
Later, I tried to retrieve information on my Ikea account about a hanging planter I bought. It reported that I couldn’t log into my Ikea account because it got stuck in a looping security check. Elsewhere on the Internet, the neighborhood teriyaki place wouldn’t let Dot access its order page. Something about Dot’s footprint seems to cause her to stumble through security checkpoints more often than Muse or Instinct; Dot offered a link to OpenAI’s page on websites that block your cloud browser traffic as an explanation. Whatever the reason, it required more manual input and browser takeovers than other AI agents, possibly because it wasn’t designed specifically to order teriyaki or handle your Ikea returns.
I really saw Dot come into her element when I gave her access to something that was completely under my control: my personal website. It’s embarrassingly outdated and I asked Dot to create a new, more stylized design. It did, and when I sat down to write down some changes I wanted made for the next version, I thought, “Screw this. I’ll just call it.” I went through the changes I wanted section by section for 10 minutes, walked away to wash some dishes, and received a ping that the next iteration was ready for me to review. Not perfect, but better. Again, AI is most useful when you give it a large set of data (me rambling for 10 minutes) and it can turn into something useful. I went through a few more rounds of iterations chatting with Dot in a browser window on my computer and using voice-to-text in the phone app while I walked around the house clearing Brio and busy train tracks.
Something about Dot’s footprint seems to cause her to stumble through security checkpoints more often than Muse or Instinct.
Dot made a lot more sense once I gave him access to my computer. There, I gave him two tasks: gather a handful of video files on my desktop featuring a TV hit I made into a clip sized for social media, and implement the changes to my personal website that we had worked on a prototype. I had to give it a surprising amount of permissions, but it worked. I watched him click into the backend of the website and recreate the prototype block by block. I stepped away from the computer and got updates from the ChatGPT app on my phone – the consolidated video clip was ready, which I reviewed over coffee. It was the most Silicon Valley I have ever felt in my life.
I absolutely could have figured out how to put all those clips together myself. I actually did it the first time, but it looked weird and was too big to upload to any social media website. Dot did a much better job and I had a clip to share within minutes. Also handled updating the website. The new design is basic, but it’s miles ahead of the complicated and outdated version I’d been using until yesterday. You have to hand it to Dotty McDotface.
Dot makes a lot more sense when you think about it and use it as “Codex, but for regular people”: a cute little blob that can take care of one part of your business while you do other things. While more personal AI agents like Muse feel like a glorified autocomplete and universal online shopping cart, Dot feels like a harbinger of the future of work.
EITHER to future of work. Presumably, since you pay up front, you won’t be subject to the kind of monetization pressure that Muse and Instinct eventually will be, which makes sense for enterprise software. Nobody wants AI to bother them while they work. That is, assuming you do the kind of work that AI can help you with and you trust OpenAI to do it. Big warnings!
He helped me with some tasks that, if I’m honest, I simply wouldn’t have been able to do. That’s pretty useful. But I don’t think $100 a month is useful, at least in my line of work. I can see how it would be much more useful to someone like our friend Casey Newton, who is in the midst of launching a new business.
And with a little help logging into my Uber Eats account, I ordered my teriyaki at the end.
