A Platform Shift, Not Just New Features
At its DevDay 2026 developer conference on September 29, OpenAI unveiled Dots, an autonomous agent powered by GPT-6 Astra, and launched the OpenAI Marketplace for enterprise customers. The announcements also covered upgrades to the ChatGPT plugin system, account sign-in, third-party app integration, and developer review workflows. Taken together, the signals point to one goal: transforming ChatGPT, which serves roughly 1.2 billion weekly active users, from a chat interface into a full platform for discovering, running, and distributing software.
Dots: From Answering Questions to Doing Work
Dots is the centerpiece of the release. Each agent runs on its own cloud computer and browser, connects to applications the user authorizes, and keeps executing tasks in the background. Unlike a conventional chatbot that must be re-briefed in every conversation, Dots can hold long-term goals, move projects forward, track dependencies, monitor issues, and conduct read-only background research even when the user is not actively chatting.
The agent already connects to an ecosystem of more than 4,000 applications and can be reached from ChatGPT, Slack, and Microsoft Teams. Autonomy comes with guardrails: users define which apps an agent may access, which actions require approval, and which are forbidden entirely. Access is rolling out gradually to eligible Pro, Business Premium, and Enterprise accounts.
ChatGPT as an AI-Native Application Platform
Plugins have historically been thin API calls behind the scenes. OpenAI has promoted them into something closer to real applications: developers can now build sidebars, interactive panels, editors, dashboards, and even complete workspaces inside ChatGPT. In demos, design tools from major creative software vendors were embedded directly in the chat interface, letting users review files, make edits, and handle comments without leaving the conversation. ChatGPT can also suggest relevant plugins based on the task at hand.
Sign in With ChatGPT
Another notable change lets users log into third-party products with their ChatGPT account. Eligible Plus and Pro subscribers can carry their existing ChatGPT usage allowances into partner apps, which lowers trial barriers for users and shifts some model-call costs away from developers. Roughly sixteen platforms joined the initial cohort.
The Enterprise Marketplace Model
The new OpenAI Marketplace lists more than thirty initial partners across customer management, security, design, and IT service categories. Its mechanic differs from a classic app store: enterprises can redirect part of the budget they have already committed to OpenAI toward purchasing partner software. In effect, OpenAI is no longer only selling its own models and subscriptions; it is channeling corporate spend into its ecosystem. That creates a new distribution channel for developers and may simplify multi-vendor procurement for buyers. A full revenue-share and billing framework comparable to established mobile app stores has not yet been published.
Developer Infrastructure Goes Store-Like
OpenAI also upgraded the tooling behind the scenes. Plugin review is now more transparent, with progress tracking, issue submission, and manual appeal options, and updates no longer require restarting the review process from scratch. The Agents API gained computer-use capabilities, enabling agents that operate software interfaces for testing, automation, and cross-app workflows. Codex moved further into the cloud, so long-running tasks continue even when a local machine is switched off.
Why It Matters: Four Unified Layers
The release can be read as one coherent distribution network:
Identity layer: a single account and subscription allowance that travels across apps.
Entry layer: ChatGPT becomes the main surface for finding and launching software.
Execution layer: cloud computers let agents act across applications.
Commerce layer: the marketplace routes enterprise budgets into third-party products.
This does not mean incumbent mobile app stores will disappear. App downloads, native clients, operating-system permissions, and procurement systems remain complex. But OpenAI is clearly building an AI-native distribution path that bypasses traditional stores. The interaction model may shift from opening an app and clicking through menus toward stating an intent and letting an agent choose the tools. For developers, understanding this new distribution logic early is likely where the next wave of AI application upside will be found.
