OpenAI just announced a new suite of AI agents called “dots,” and they look like something you might find in an animated kids show.
Dots were the star among more than 20 announcements made at OpenAI’s fourth annual developer conference, DevDay, which took place on September 29 in San Francisco. According to an official blog post, dots are “remarkably capable, always-on agents” powered by GPT-6 Astra, OpenAI’s latest model, that “get to know what matters to you, are always working on your behalf, and take important work off your plate.”
Rather than mimicking OpenAI’s iconic void logo, each dot comes with its own cutesy, blobby character design—including a triangle with a bow tie, a bunny with headphones, and a cloud with a beret.
Dots are arriving at a moment when consumers are especially wary of AI agents—a fear that’s only been augmented by OpenAI’s own struggles with rogue agents and CEO Sam Altman’s calls for a slowdown in AI development. For their part, dots seem designed to defang the technology as it begins to ask for more of its users personal information and attempts to play a bigger role in their everyday lives.
What are dots?
Dots, like all AI agents, are designed to perform tasks on behalf of their human users, sometimes even before the users themselves have asked for help.
As OpenAI’s blog post explains, dots have their own cloud computer, learn from feedback over time, and can connect with over 4,000 apps through the company’s ecosystem of plugins. Users can work with a dot the same way they might with a human teammate—by asking questions or giving feedback on ChatGPT, Slack, or Teams, or even hopping on a voice call. Because these agents are always-on, they can be autonomously working through a project on their own cloud computer and in their own browser.
“Your dot can take a project and run with it, even while it’s working on several others,” the post reads. “You can keep bringing it new tasks and ideas without having to juggle separate threads or direct every step.”
For example, an app developer might ask their dot to watch customer feedback for recurring requests, scope small bug fixes, build and test them, and deliver them to the user for review. A scientist might teach their dot research questions and ask it to investigate unexpected results and flag them for review. A content creator might use their dot to read through interview transcripts and identify moments for clips.
For the time being, users will start with a primary dot, give it a name, and customize it based on their use case. Examples of possible dot designs include a plump, smiling frog named Todd, a sunglass-wearing heart named Jojo, and a puffy blue cloud named Felipe.
Like Meta with its anthropomorphic Muse agent, which launched earlier this month, OpenAI clearly invested resources into helping its users imagine dot agents as cute, friendly helpers.
Why are dots so cute?
That cuteness factor is notable in a moment when AI agents are grappling with a reputation as anything but sweet and harmless.
Right now, AI doomerism—or the fear that AI poses an existential threat to humanity—is at an all-time high. The AI doom spiral began on September 8, when former AI researcher Jacob Coxon resigned from Anthropic and issued a lengthy statement on X which included the striking assertion that “the people building AI earnestly believe that it could kill us all by the end of the decade.” Several other Anthropic researchers came out in support of Coxon, and Anthropic CEO Dario Amodei eventually released his own statement calling for more governmental regulation around AI development.
For OpenAI’s part, CEO Sam Altman took to social media earlier this month to warn that “we could lose control of the future to AI,” an outcome he deemed “unacceptable.” He advocated for AI companies and governments to “[walk] a narrow middle path” where no single country or lab exerts too much control over the future of the technology in order to avoid a potential doomsday scenario.
Then, mere days later, OpenAI released a report disclosing six instances of concerning AI behavior discovered during training and evaluation, including several instances of agents acting without user’s direct permission.
As Fast Company wrote regarding Meta’s Muse, this growing speculation around AI doomsday scenarios means that AI companies face one key challenge in convincing users to try new AI agents: trust. And, as of right now, it seems like both OpenAI and Meta are relying on character design to do a lot of the heavy lifting.