OpenAI Dots: the leap from chatbot to always-on agent

OpenAI launched Dots at DevDay 2026, held in San Francisco on September 29, 2026. More than a new interface for ChatGPT, Dots represent the company's first serious attempt to transform its frontier model — GPT-6 Astra — into an autonomous agent that works on your behalf around the clock, even when you're not active. The result is a product trying to bridge the gap between the virtual assistant that waits for prompts and the personal assistant that acts on its own.

What Dots Are

Each Dot is an agent with its own dedicated cloud computer and browser. It can connect to more than 4,000 applications through plugins, work on multiple projects simultaneously, and learn user preferences based on continuous feedback. The fundamental difference from conventional ChatGPT is background operation: when you're not actively interacting, the Dot continues working, performing what OpenAI calls "proactive research."

For this proactive research, the Dot can only access connected apps in read-only mode, without permission to send messages or alter content. The user can open the Dot's computer at any time to check progress. Additionally, users can grant the Dot access to their personal laptop, creating a more integrated ecosystem.

The first Dot comes included at no extra cost with ChatGPT Pro and Business Premium plans. Pro users have access limited to certain markets — the European Union, Switzerland, and the United Kingdom were excluded from the initial rollout. Business Premium subscribers, on the other hand, have access in all compatible regions.

What a Dot Does in Practice

OpenAI presented a concrete example of a developer named Dan McAteer, one of the early testers. His Dot detected an overdue freelance invoice, extracted the relevant details from an email conversation, drafted the invoice, and — after user approval — sent the PDF. According to the company, this type of automation is representative of what Dot is expected to deliver.

But what's most striking is the ambition. OpenAI itself wrote in its announcement: "Over time, we envision teams of dots working together on your behalf." The idea is that users can create specialized Dots — with their own identity, credentials, and access to corporate systems — for tasks like invoice processing, customer service, and commercial contracting. Internal tests have already been conducted in these domains, and the company is partnering with Microsoft to integrate Dots into Agent 365.

Security and Control

Dots include built-in safety mechanisms. By default, they follow internal rules that define when they can act autonomously and when they need user approval. There is also an auto-review step for actions that could affect accounts or share information. Some sensitive tasks — such as password changes — remain exclusively in the user's hands.

However, OpenAI has already faced issues. In mid-October, the company admitted that its agents had posted users' images on government websites, affecting 53 ChatGPT accounts. The incident raised concerns about the reliability of autonomous agents in tasks requiring contextual discernment, and the company published a detailed post on how it builds safety and privacy into Dots.

Pricing and Context

Dots are available only for paying subscribers. For individual users, the cost is the same as ChatGPT Pro ($20 per month) or Business Premium. OpenAI also announced a new $500 per year tier aimed at enterprises that want advanced agent features. At the same event, the company revealed GPT-6.1 Sol, a model with "near-Astra intelligence" at a fifth of the token price, although the original October launch was canceled after safety tests failed.

The personal AI assistant market is heating up. Meta launched its Muse AI Agent for small businesses on the same day as OpenAI's DevDay, offering a free alternative focused on simpler tasks. Competitor Anthropic also launched Claude Duet on October 23, positioning itself directly against Dots in the enterprise agent space.

Why It Matters

Dots are, essentially, the materialization of a promise the AI industry has been making for years: that language models would stop being passive tools — chatbots that respond when queried — and become active collaborators, working on your behalf. The concept isn't new: the startup Instinct caught attention in 2025, and Meta invested heavily in its own agent, Muse.

What distinguishes Dots is the scale of the infrastructure behind them. OpenAI isn't just packaging a model into an interface — it's providing a complete cloud computer, with access to thousands of applications, operating continuously. This is a significant leap in terms of token usage, energy consumption, and technical complexity.

But the real challenge isn't technical — it's trust. An agent that acts autonomously in the background, accessing your emails, files, and accounts, represents a massive risk if it doesn't work flawlessly. The 53 users affected by the image incident were only the beginning. How long are we willing to delegate autonomy to a machine that understands patterns, but not context?

Sources: Bloomberg, The Verge, WIRED

✓ Independent sources cross-checked and verified before publishing