OpenAI released its autonomous agent Dots on DevDay 2026 and officially launched the OpenAI Marketplace, an enterprise application marketplace. At the same time, ChatGPT's plugin system, identity login, third-party application access, and developer review mechanism have also been upgraded.
The most crucial signal of this release is not "a few more features", but that OpenAI is transforming ChatGPT, which has approximately 1.2 billion weekly active users, from a chat portal into a comprehensive platform for software discovery, invocation, operation, and distribution.
Dots: From 'you ask it to answer' to 'it does it fyou'
Dots is OpenAI's most anticipated product this time. It is driven by GPT-6 Astra, with independent cloud computing and browsers, capable of connecting user authorized applications and continuously executing tasks in the background.
Unlike traditional chatbots, Dots does not require users to provide background information every time. It can understand long-term goals, proactively advance projects, follow up on dependencies, monitor issues, and even conduct background research in read-only mode when users do not have direct conversations.
At present, Dots has connected over 4000 application ecosystems and can be used in portals such as ChatGPT, Slack, Microsoft Teams, etc. However, it does not have unlimited autonomous actions: users need to authorize access scope and can also set which operations must be approved and which operations are prohibited from being executed.
At present, Dots is gradually opening up to eligible ChatGPT Pro, Business Premium, and Enterprise users.
ChatGPT is becoming an 'AI native application platform'
In the past, plug-ins were more like "calling an interface behind". OpenAI has upgraded it to something more like an application this time.
Developers can build sidebar entrances, interactive panels, editors, dashboards, and even complete workspaces within ChatGPT. Users can directly operate third-party applications in ChatGPT during conversations without frequent jumping.
In the live demonstration, applications such as Figma and Adobe were directly embedded into the ChatGPT interface, allowing users to view design drafts, modify content, and process comments. At the same time, ChatGPT can automatically recommend plugins or applications that may be helpful based on the user's current task.
Another important change is the use of ChatGPT login. Users can use their ChatGPT account to log in to third-party products, and eligible Plus and Pro users can also bring their ChatGPT usage quota to partner applications.
The first batch of partners includes about 16 platforms such as Devin, Notion, Vercel, OpenClaw, etc. This is very attractive to developers: some model call costs no longer have to be borne by developers, and users can also try third-party AI products at a lower threshold.
Enterprise application market: Use OpenAI quota to buy third-party software
OpenAI has also launched the OpenAI Marketplace, an enterprise application marketplace, with over 30 initial partners including Salesforce, HubSpot, CrowdStrike, Figma, Adobe, ServiceNow, and others.
Its logic is quite different from traditional app stores: enterprise customers can use a portion of the promised purchase amount to OpenAI to purchase software from partners. That is to say, OpenAI is not only selling its own models and subscriptions, but also beginning to channel enterprise budgets towards the products of its ecosystem partners.
This means new distribution opportunities for developers; For corporate clients, it may reduce the complexity of multi supplier procurement.
However, OpenAI has not yet announced a complete revenue sharing or billing system similar to that of Apple and Google Play Store.
The developer ecosystem is also becoming "app store oriented"
In addition to changes aimed at users and enterprises, OpenAI has also upgraded its developer infrastructure.
The plugin review process has become more transparent, allowing developers to view the review progress, submit issues, apply for manual review, and avoid having to re review from scratch every time the tool is updated. At the same time, the Agents API has added Computer Use capability, allowing developers to build intelligent agents that can operate software interfaces for testing, automated processes, or cross application tasks.
Codex is also further cloudification. Developers can run long tasks in the cloud, even if the local device is turned off, the tasks can still continue to be executed. These capabilities together constitute the infrastructure layer of OpenAI for the Agent era.
Why is this matter worth taking seriously?
OpenAI is not upgrading the model at a single point this time, but building a more complete distribution network:
Unified identity layer: Sign in with ChatGPT allows account and subscription limits to flow across applications;
Unified entrance layer: ChatGPT becomes the main interface for software discovery and invocation;
Unified execution layer: Dots has cloud based computers that can execute tasks across applications;
Unified business layer: Marketplace allows enterprise budgets to enter third-party software ecosystems.
This does not mean that the app stores of Apple and Google will be immediately replaced. Mobile downloads, native applications, system permissions, and enterprise procurement systems are still complex. But OpenAI has clearly attempted to establish an AI native distribution path that bypasses traditional app stores.
In the future, software interaction may further shift from "opening the app - clicking on the menu - completing the operation" to "expressing intention - intelligent agent selection tool - automatic execution". For developers, the earlier they understand this new distribution logic, the more likely they are to seize the next wave of AI application dividends.
