OpenAI turns ChatGPT into an app and agent platform
ChatGPT will integrate services such as Spotify, Canva, and Zillow into its conversations. OpenAI has also unveiled AgentKit and opened GPT-5 Pro and Sora 2 to developers.
OpenAI wants ChatGPT to be more than just a chatbot and become the gateway to other digital services. The company today unveiled apps integrated into conversations, a toolkit for building AI agents, and API access to GPT-5 Pro and Sora 2.
The push comes with an advantage that is hard to ignore: ChatGPT has reached 800 million weekly active users, Sam Altman announced during DevDay 2025. That is 100 million more than the figure reported in August, although it does not represent paid subscribers.
Spotify, Canva, and Zillow enter the conversation
Users will be able to ask ChatGPT to use an app simply by mentioning it in a message. Spotify can help create a playlist; Canva can turn an idea into a design; and Zillow can show homes that meet specific criteria.
The initial partners also include Booking.com, Coursera, Expedia, and Figma. The result combines conversation with interactive interfaces: maps, designs, videos, or listings can appear directly inside the chat, without forcing users to start over in another app.
To build these integrations, OpenAI has launched the Apps SDK in a preliminary phase. The toolkit is built on MCP, short for Model Context Protocol, an open protocol originally created by Anthropic to connect AI models with external data and tools. Its adoption reduces the need to create a separate connection for each assistant.
The idea recalls the plugins ChatGPT introduced in 2023 and later retired in favor of custom GPTs. The difference is that these new apps can combine actions, data, and visual interfaces directly inside a conversation. OpenAI is not merely hosting small specialized chatbots; it is seeking to become a distribution layer for third-party services.
The 800 million weekly users explain the appeal for companies such as Spotify and Canva. If some searches, bookings, and purchases begin in ChatGPT, appearing prominently inside the assistant could become nearly as important as visibility on Google, app stores, or mobile operating systems.
AgentKit takes agents from prototype to production
The second major announcement is AgentKit, a set of tools for building and deploying AI agents. An agent does more than answer a question: it can chain together steps, consult sources, use external software, and carry out a task with some degree of autonomy.
AgentKit brings together four main components:
- Agent Builder, a visual editor for designing an agent’s workflow, decisions, and available tools.
- ChatKit, a conversational interface that companies can embed in their own products and adapt to their brand.
- Evals for Agents, tools for checking performance step by step, evaluating components, and optimizing instructions.
- Connector Registry, a managed registry for linking agents to internal systems and third-party services while maintaining centralized controls.
During the event, OpenAI demonstrated the creation of a workflow and two agents in under eight minutes. The demo shows that the initial setup can be streamlined, but it does not solve the hardest problem by itself: ensuring that the system behaves correctly when faced with incomplete data, malicious instructions, or situations that did not arise during testing.
Evaluations are especially important because agents can have consequences beyond the chat. An error in drafting a response is annoying; an error that modifies a database, sends an email, or processes a refund can carry financial and legal costs.
GPT-5 Pro and Sora 2 come to the API
OpenAI has also added GPT-5 Pro and Sora 2 to its developer platform. The API allows companies to integrate the models into their own products without directly using the ChatGPT interface.
GPT-5 Pro is aimed at tasks that require more reasoning and computing power than the standard version. Sora 2 expands the offering into video generation, opening up uses in advertising, entertainment, training, and audiovisual prototyping.
API availability matters more than a new option inside ChatGPT: it allows other companies to build commercial services on top of these models. It also increases their dependence on OpenAI for factors such as pricing, usage limits, availability, and content policies.
ChatGPT competes to become software’s interface
The DevDay announcements point to two complementary moves. With Apps SDK, OpenAI is bringing external services into ChatGPT. With AgentKit, it is trying to get companies to build their own agents on OpenAI’s infrastructure.
That brings ChatGPT closer to the role of a platform, even though it is not yet an operating system in the strict sense. The comparison is useful because it brings together users, developers, apps, and connection mechanisms; but it lacks the direct control over devices held by Apple, Google, or Microsoft.
The immediate challenge will be governing what data each app receives and what actions it can perform. The more useful ChatGPT becomes as an intermediary, the more information will have to flow among the user, OpenAI, and external providers. Understandable permissions, confirmation for sensitive operations, and activity logs will determine whether this platform is used for more than flashy demonstrations.
