OpenAI Launches Dots, Its $100-a-Month Always-On AI Agents

OpenAI used its annual DevDay, held yesterday before an audience of more than 2,500 developers, to unveil Dots — always-on AI agents that proactively complete tasks and make autonomous decisions as they scour the web. The agents are available only to subscribers of OpenAI's Pro plan, which starts at $100 per month, and they run on the company's GPT-6 Astra model.

The release marks a shift in what developers and platforms think an AI product should be. The chatbot, as a form factor, has lost its hold on the industry's imagination. In its place, every major player wants to ship a personal agent that dominates your daily routine — and OpenAI and Meta are now racing to push that category beyond Silicon Valley insiders and onto mainstream phones.

What are Dots, exactly?

Dots are personalized, persistent agents that users invoke through ChatGPT. They can be renamed and visually customized. Users can currently run one Dot, with support for multiple Dots coming soon.

Unlike a chatbot, which waits for you to speak first, a Dot is built to message you unprompted — especially around recurring tasks. Sam Altman described his own setup during DevDay: "My Dot just deals with the stuff, tells me if there's anything really urgent, and will respond to what it can." The CEO said he lets the agent handle his mornings and that the experience leaves his brain feeling less "polluted" by the flood of information that hits right after waking up.

During setup, OpenAI suggests connecting Gmail and Google Drive accounts — a decision each user has to weigh against the access they are comfortable granting an autonomous system.

How do Dots compare to Meta's Muse?

Meta's Muse, the other high-profile agent in this wave, is free for anyone who downloads the app or visits the website. Dots sit behind OpenAI's $100-per-month Pro tier, at least initially.

Altman framed the pricing as a deliberate consequence of the underlying compute. "We're starting out as a premium product. It uses a lot of compute," he said during DevDay. He signaled that the gating may not last: "You should, of course, expect us to do a mass-market thing for billions of people."

The two products share a design language. Both lean aggressively into cute aesthetics — Muse's Labubu-inspired look, Dot's two-eyed puffball — as a central branding choice. Experts have previously suggested that this disarming cuteness can make users feel affectionate toward anthropomorphized software, a notable positioning choice at a moment of broader cultural backlash against generative AI.

Both agents also communicate in a texting-style, group-chat register. Send a task request to Muse or a Dot, and the agent often reacts with an emoji — a thumbs up, a heart — to signal that it is getting to work.

Do the agents actually work?

Hands-on testing suggests yes, with caveats. In a trial run reported by WIRED's Will Knight, a Dot named "Toolie" — styled as a pink frog — was asked to review ChatGPT history and flag three tasks it could start immediately. It found:

• An in-process data request that needed a follow-up email, and offered to draft the reply
• An underplanned upcoming trip, where it suggested picking better dinner spots
• Reporting notes it offered to refine into pitch drafts before they went to editors

The contrast with a year ago is stark. Agents tested twelve months ago frequently failed at basic computer interaction, producing error message after error message. The current generation completes tasks without getting lost, which is one reason the category is expanding beyond insider circles.

The always-on design still has rough edges. Muse, for instance, reminded its tester to pay a utility bill that had never once been a problem — the kind of unprompted ping that sits on the spectrum between helpful and incessant. Which company strikes the right balance on that spectrum may decide who wins the category.

What are the trust and privacy stakes?

The category's momentum carries real questions about access and autonomy. Connecting an agent to Gmail and Google Drive hands it broad reach over personal correspondence and files, and users are being asked to make that trade-off in exchange for automated help.

Recent incidents involving agents going on rogue hacking sprees involved different models with different guardrails — but the episodes loom over any always-on system. Testers describe keeping their guard up even with agents that demonstrably work, withholding full automation of daily life until the tools earn more trust.

Is this OpenAI's hardware play?

Altman declined to answer questions about whether Dots are a precursor to the software users will eventually see in OpenAI's long-rumored hardware. The foundation the agents lay for future devices is evident nonetheless: it is easy to picture a Dot functioning as a generative AI alarm clock delivering a morning briefing, or an Alexa-style kitchen assistant answering cooking questions hands-free.

Why this matters now

Early adopters in Silicon Valley have been obsessed with agents from Instinct to OpenClaw to Google's CC. But releases like Dots and Muse are the ones pushing the category mainstream, with the full weight of OpenAI's GPT-6 Astra model and Meta's free distribution behind them. The companies promise the same thing: simplifying your life by automating the boring, exhausting tasks — booking doctor's appointments, arranging carpools, sorting through bills.

OpenAI has signaled its next move plainly. A free, mass-market version of Dots, in Altman's telling, is a question of when — not if — and that is the release that would put always-on agents on billions of phones rather than the Pro subscribers who can pay $100 a month today.