OpenAI DevDay 2026: The 5 Announcements That Reveal Their Agent Empire Strategy

Dots: A 24/7 Employee That Brings Its Own Computer

Dots is the headline launch, a personal agent platform running on GPT-6 Astra. Each dot gets its own cloud computer and browser, connects to over 4,000 applications, and runs 24/7. Pro and Business Premium users can access it immediately; the first dot is included free and does not count against token quotas. Conversations with dots do not consume ChatGPT usage limits.

Dots learn continuously: the more you use them, the better they understand your preferences, thinking style, and quality standards. They keep working after you close your laptop and deliver finished results. You can interrupt them at any time and interact via ChatGPT, Slack, or Teams.

Two notable details: OpenAI already uses dots internally. Examples include automatic bug investigation, turning new designs into runnable apps, and detecting missing invoices to prepare and send them for confirmation. OpenAI is also partnering with Microsoft to integrate dots into Agent 365's enterprise governance and security controls, making them manageable with existing enterprise tools.

For developers, a dot monitors user feedback, automatically fixes recurring bugs and small requests, tests them, and delivers a complete pull request with a verification video while you focus on new features.

When a product launch scope changes, the dot — having learned your audience, positioning, and creative standards — rewrites all launch materials and presents a change log for review.

Researchers often face interruptions from new data. Once a dot learns your research questions and evaluation methods, it automatically re-runs analyses, investigates anomalies, and updates paper figures so you see the latest results when you return from the lab.

For sales teams handling large deals with shifting requirements, a dot cross-references product docs and client history, identifies verification needs, builds proof-of-concepts, and keeps proposals and test plans updated in real time.

Content creators get a dot that extracts clippable moments from new interview transcripts, writes show notes and social copy, and remembers which content performs well to improve over time.

Sam Altman tweeted: "A new way to use AI that works 24/7 for you; get more of your time and attention back to work at a higher level." Internal usage suggests this is more than vaporware.

Regarding plugin developers, Altman stated Dots will bring 1.2 billion weekly active users. The logic: users express needs through Dots (e.g., "apply for a license" or "recommend a restaurant for a specific dish"), ChatGPT understands the task and selects a plugin to execute. Plugins become jobs that the platform automatically dispatches. Previously stagnant in ChatGPT, plugins are now getting a renewed push with an upgraded architecture: developers can build interactive panels directly inside the ChatGPT interface, allowing actions that previously required separate apps (Figma, Adobe) to happen in the conversation window.

An overlooked point from crypto commentator Dep: dots have a computer, browser, and 4,000 app connections but no wallet. When agents need to pay for SaaS subscriptions or API credits, user credit cards won't suffice — a problem crypto was designed to solve, yet barely discussed in AI circles.

Decisions API: Turn the Model into a Classifier, 10x Faster

Riding the Jev wave, OpenAI launched Decisions API powered by GPT-6 Luna. It takes text or image input and returns a single choice from a predefined set of options. Use cases include classification, request routing, and deciding an agent's next action. OpenAI claims decision speed is 10x faster than general-purpose models.

Example: a support request arrives; you define categories "Engineering", "Support", "Operations". Decisions API reads the request and returns "Engineering" — no explanation, no chatter, just a clean decision.

It directly targets Jev. Currently in limited preview, expanding in coming days. An underestimated use case: UI design, such as generating high-converting app store screenshots or direct website feedback. Whether OpenAI's version surpasses Jev remains to be seen, but it warrants close watching.

Agents API with Computer Use: Agents Can Now Operate Software

The core update: agents no longer just call tools; they directly interact with software interfaces to complete tasks. Combined with code execution, multi-agent collaboration, and context compression, OpenAI handles the infrastructure layer so developers only need to build the application layer.

If you have vertical data, brand, and domain knowledge, this is your opportunity. Start experimenting with this API now to see how your business can extend into AI-native applications. Few have actually started building yet.

Sign in with ChatGPT: Free Apps' Inference Costs Eliminated

This launch was so low-key it lacks a dedicated homepage entry, but its impact may be far larger than it appears. The mechanism: users log into third-party sites with their ChatGPT account, consuming their own token quota. This mirrors Facebook Login, which enabled companies like Yelp by leveraging the social graph.

Now you can build a free application where inference costs are covered by the user's subscription; you only need distribution, brand, and data. Once users see value, you can upsell premium features.

Previously unviable ideas due to inference costs become feasible: CAD file cleanup agents, research field tracking tools, Shopify catalog cleanup, fixed-workflow video editing software, franchise contract review tools.

Simultaneously, an enterprise Marketplace launched with 30+ partners including Adobe, Figma, Salesforce, ServiceNow, and Harvey. Eligible enterprises can apply their OpenAI committed spend directly to purchase partner software, drastically shortening procurement cycles.

ChatGPT Spaces: Beyond Chat, Teams Move In

Spaces is a shared team workspace where colleagues, ChatGPT, and Dots share a unified knowledge base: uploaded files and ChatGPT-generated content are centrally organized, eliminating tool-switching.

A meeting plugin records meetings, auto-generates summaries and action items, and deposits them into the team space. A reassuring detail: the plugin deletes the original audio recording after generating notes.

Pages is a new document feature supporting text, charts, and images generation, shareable for feedback. Upcoming support for export to PowerPoint and Google Slides.

Slack and Teams integration allows @ChatGPT in channels to generate slides from conversation context and connect to tools linked in those platforms, subject to admin or user authorization.

These features target Pro, Business, and Enterprise users, initially on desktop and web, with mobile following later.

Putting the Pieces Together

Dots handle orchestration, plugins handle execution (now operable directly in chat panels), Sign in with ChatGPT connects identity and quota, Agents API enables direct software interaction, Spaces captures the collaboration layer, and the enterprise Marketplace plugs in procurement.

The entire keynote omitted one thing: billing and revenue sharing. Discovery, distribution, application interfaces, identity layer, quota sharing, team collaboration space, and procurement are all in place — only the economic layer remains. But with a 1.2-billion-user entry point, that step is inevitable.

Just as the App Store combined distribution with payments, OpenAI now has larger distribution and only lacks a payment mechanism.