OpenAI Launches Dots, Always-On AI Agents, at DevDay 2026

OpenAI announced Dots, always-on AI agents that complete tasks on a user's behalf, at its DevDay 2026 event on Tuesday in San Francisco. The rollout starts today for subscribers to ChatGPT's Pro tier, which costs $100 a month, positioning OpenAI in direct competition with Meta's Muse agent, which topped smartphone app download charts after its recent release.

Dots are powered by OpenAI's GPT-6 Astra model. The company depicts them as cute blobs that users can personalize and assign multi-step projects to complete. Their defining feature is the "always-on" design. Unlike a single-turn chatbot experience, Dots constantly crawl the web and keep working on whatever task a user assigns them. They also pull context from connected apps and learn more about a user's preferences over time.

How the agents behave in practice

The distinction between a chatbot and an always-on agent shows up in the demos OpenAI shared at DevDay.

In one example, a Dot helped a user plan dinner. The Dot proactively noticed that the user's calendar showed they were working through dinner, then messaged with two GrubHub food delivery options along with pricing for each. The user replied with their food pick and when they wanted it ordered. The agent initiated the interaction — the user did not ask for restaurant suggestions.

In a second example, a user collaborated with their agent to launch a new website. That demo points to a broader ambition than lifestyle assistance: multi-step project work, the territory that has so far been the domain of both consumer agents and developer-oriented automation tools.

The dinner example illustrates the architectural shift. Dots monitor connected services continuously, not only when prompted. That constant monitoring, combined with persistent memory of user preferences, is what separates the product from ChatGPT's conversational model, where context resets or degrades between sessions.

Where you can reach a Dot

Users can message Dots through ChatGPT, Slack, and Microsoft Teams, with the user's context shared across those modes. A Dot that knows your calendar in ChatGPT will know it in Slack too. Pro users can also join a waitlist to text with their Dots in iMessage or RCS messaging on Android — a notable move because it puts an OpenAI agent inside native phone messaging channels rather than confining it to OpenAI's own apps.

The iMessage and RCS integration exists only as a waitlist at launch, signaling capacity constraints or a staged rollout. OpenAI has not said when messaging access will open broadly.

Pricing and availability strategy

The launch follows a pattern OpenAI has used repeatedly: ship new features first to the Pro tier, then expand to a larger subset of users over time. At launch, users can control one Dot. The company is expected to eventually roll out the ability to control multiple agents at the same time, though OpenAI gave no date for that expansion.

The single-agent limit matters for the product's trajectory. Multi-agent orchestration — running several Dots in parallel on different projects — is where always-on agents could compound in value, but it also multiplies the risk surface of autonomous software acting on a user's accounts.

Guardrails and user control

Dots are designed to get explicit approval before taking more sensitive actions on a user's behalf, like installing software or changing a password. That requirement addresses one of the central concerns around autonomous agents: software that can act on accounts and systems needs friction at the moments where mistakes or misuse carry real cost.

Users can also enable a Custom Rules tool to set explicit boundaries they don't want agents to cross, as well as tasks that need more direct permission to complete. The Custom Rules tool effectively gives users a policy layer over their agent — a way to define, in advance, what the Dot may never do and what requires a confirmation prompt each time.

The combination of approval gates and custom boundaries places OpenAI on the more conservative end of the agent autonomy spectrum at launch. Whether that balance holds as Dots gain capabilities — and as users push them toward longer, more complex workflows — will be a key question for the product's evolution.

The market context: a breakthrough moment for personal agents

Personal agents are having a breakthrough moment. The year started with early adopters in Silicon Valley latching onto tools like OpenClaw. The recent release of Meta's Muse agent then reached a much larger user base, one now actively experimenting with personalized agents. Muse rocketed to the top of smartphone app download charts, establishing consumer agents as a mainstream category rather than a developer curiosity.

Dots is OpenAI's answer, and the framing in coverage of the launch is blunt: Dots has come for Meta's lunch.

The competitive stakes are structural. Meta distributes Muse through smartphone app ecosystems and can leverage its existing user base across Instagram, WhatsApp, and Facebook. OpenAI's distribution runs through ChatGPT — already one of the most-used AI products globally — plus enterprise channels like Slack and Microsoft Teams, and now native phone messaging via iMessage and RCS. The two companies are betting on different wedges into the same behavior: a persistent, personalized agent that knows the user and acts without being asked.

The $100-per-month entry point also matters. At launch, Dots are a premium product for OpenAI's highest-paying subscribers, not a mass-market play. That mirrors how OpenAI has handled prior feature launches, but it means the head-to-head battle with Muse — which reached mass download charts — will only begin when Dots reach cheaper tiers or free users. OpenAI has not announced timing for any broader rollout.

Why it matters

Three things make this launch significant beyond the product itself.

First, always-on agents represent a shift in the default interaction model with AI. If Dots work as demonstrated, users stop initiating every exchange. The agent watches, infers, and proposes. That changes both the utility and the privacy calculus of consumer AI, since constant crawling and cross-app context access are prerequisites for the behavior.

Second, the guardrail design — approval gates for sensitive actions plus Custom Rules — will be watched closely as a template for how the industry handles agent safety at consumer scale. Regulators and rivals alike will study whether explicit-approval requirements survive contact with real usage.

Third, the timing sharpens the OpenAI-Meta rivalry. Muse proved demand exists at mass-market scale. Dots arrive weeks or months later, gated behind a $100 subscription, but backed by GPT-6 Astra and OpenAI's ChatGPT distribution. The next phase of the consumer AI market — persistent agents rather than chat windows — will likely be shaped by how quickly OpenAI widens access and whether Meta's agent can match the multi-step, cross-app depth Dots promise.

For now, the product is live for Pro subscribers starting today, one Dot per user. OpenAI has not said when multi-agent control, broader tier access, or full iMessage and RCS availability will arrive. OpenAI describes this as a developing story and is encouraging users to check back for updates — which suggests more DevDay announcements, or early product changes, may land in short order.