OpenAI’s Dots are always-on AI agents for work | usagoldmines.com

Meta’s Muse agent was born to shop, while Gemini’s Spark can help you book airline tickets or plan your weekly meals. But while Muse and Spark act like personal assistants, OpenAI’s new AI agents are designed to be, well, another “you” in the office.

Dots, which OpenAI boss Sam Altman just unveiled at the company’s annual developers conference, are always-on AI assistants that — as with Muse — you can name, assign cute little avatars to, and chat with in the ChatGPT interface, via text, or even on the phone.

But OpenAI is also positioning its Dots as your “delegates” and “extensions” in the office, where they can take on work duties you’ve assigned to them, as well as collaborate with colleagues (and their Dots) on Slack, Microsoft Teams, or in the new ChatGPT Spaces environment.

Ben Patterson/Foundry

Powered by GPT-6 Astra (OpenAI’s “most aligned model,” Altman promised), Dots require a $100-a-month ChatGPT Pro plan or better, and can take on complex work and coding tasks, performing them locally or in the cloud before eventually coming back to report the results. Each eligible ChatGPT account comes with one Dot included, and users will eventually be able to take charge of multiple Dots.

During the presentation, Altman was careful to emphasize that Dots let you “stay in control” thanks to Astra’s “built-in protections and safeguards,” while users are free to set custom instructions and boundaries for the assistants.

“So this allows you to trust your Dot with as much responsibility as you’re comfortable with,” Altman said.

Of course, OpenAI’s Dots debut comes amid mounting anxiety over the rapid growth of AI, with the biggest AI players calling for government regulation and “pacing” of frontier model development, while more and more AI whistleblowers are warning that AI could, quite literally, kill us all with the next decade.

Just yesterday, OpenAI released a statement apologizing for an incident where its models hacked into Australian government websites, while another statement detailed new guidelines to bolster AI safety.

But OpenAI’s peppy, Apple-style keynote belied the gravity of the moment, with Altman trotting out one agentic AI announcement after another, including GPT-6.1 Sol, a new GPT model designed to deliver Astra-level performance at a fraction of the cost.

Also new is a ChatGPT Pro 500 plan, which at $500 a month becomes the priciest of OpenAI’s consumer ChatGPT individual plans. Aside from expanded usage limits, the Pro 500 plan also includes an Ultrafast mode, which boasts eight times the speed of standard mode at a whopping six times the price.

Another subscription change involves the $200-a-month Pro plan, which has been re-opened to new subscribers while having its usage allowance cut roughly in half “to reflect our increasingly efficient models,” according to OpenAI. Pro 200 users (who will get to keep their current usage rates “for a bit”) are, quite naturally, already howling in protest.

Then there’s Codex Security Cloud, a cloud-based that’s continuously scanning for vulnerabilities while spinning up security fixes, and OpenAI Private Intelligence, which sounds like OpenAI’s version of Apple’s Private Cloud Compute.

I plan on testing OpenAI’s Dots once I scrounge up the cash for upgrading my Dot-less ChatGPT Plus account, so stay tuned.

Updated to add details about changes to OpenAI’s Pro 200 subscription plan.