What did OpenAI announce at DevDay 2025, and what does it mean for you?

Yesterday, Monday, October 6, 2025, OpenAI held DevDay 2025, its yearly conference for software developers, at Fort Mason in San Francisco. CEO Sam Altman opened the keynote with a few numbers: 800 million people use ChatGPT each week, 4 million developers build on OpenAI's platform, and its API processes 8 billion tokens per minute. An API, or application programming interface, is the way other programs talk to OpenAI's models; a token is a small chunk of text.

The announcements fall into three groups: apps inside ChatGPT, tools for building agents, and new models for developers.

What OpenAI announced

Smaller items included a dashboard showing the health of OpenAI's services and faster responses for GPT-5 on its priority processing tier.

Apps inside ChatGPT

The headline idea is that ChatGPT is becoming somewhere other software runs. You can start a message with an app's name, for example "Spotify, make a playlist for my party this Friday," and the app appears inside the chat with its own interactive interface. ChatGPT may also suggest an app when it fits, such as Zillow's map of homes when you talk about moving.

TechCrunch notes the difference from OpenAI's earlier GPT Store: that was a separate catalog, while these apps show up right in ChatGPT's answers. Apps are built on the Model Context Protocol (MCP), an open standard for connecting AI to outside tools and data. On stage, OpenAI showed Canva producing four poster designs for a dog-walking business inside ChatGPT, and a Coursera lecture that ChatGPT could explain as it played.

Building agents faster

An agent is an AI system that works toward a goal over several steps, using tools along the way. AgentKit gathers the pieces companies need to build one. To prove the point, OpenAI's Christina Huang designed, tested and published a working agent on stage in under eight minutes, with 49 seconds left on the clock, according to CNBC.

Codex, meanwhile, is now generally available. Romain Huet, who leads developer experience, used Codex and his voice to build controls for a stage camera and lights without typing code. Altman summed up the theme: software that used to take months can now be built in minutes by someone with a good idea.

A worked example: what each piece is for

To see how the announcements fit together, imagine a small online bookstore:

1. Apps SDK: the store builds a ChatGPT app, so a reader can type "Bookstore, find me a mystery under $20" and browse covers right in the chat.
2. AgentKit: the store designs a customer-service agent in Agent Builder that checks order status and drafts replies, and adds it to its own website with ChatKit.
3. Codex: a developer asks Codex to add a "gift wrap" option to the checkout code and to write tests for it.
4. gpt-image-1-mini: the store generates simple banner images for weekly promotions at a lower cost.

Why it matters

DevDay shows OpenAI moving beyond a chatbot that answers questions toward a platform with its own apps, agents and developer tools. For users, it means fewer trips between tabs; for developers, a chance to reach ChatGPT's very large audience.

There are open questions. TechCrunch asks how much of a conversation app makers will see, and how ChatGPT will choose between competitors such as two delivery services. Markets reacted quickly: CNBC reports that shares of Figma, HubSpot and Coursera rose after they were mentioned on stage, a response Altman called strange. OpenAI also announced a chip agreement with AMD the same day, and Altman and Greg Brockman repeatedly said computing power remains a bottleneck.

The day ended with a conversation between Altman and Jony Ive, the former Apple design chief, who said he had never seen anything with capabilities like today's AI in his career. The talk was not streamed.

Which DevDay 2025 announcement should you try first?

If you simply use ChatGPT, try the apps. Start a message with the name of a launch partner such as Spotify, Canva, Coursera or Zillow, and see how the app appears inside the conversation. Connect only the apps you need, and check what you are sharing with each.

If you write code, try Codex on a small, well-defined task, and review its work before you merge it.

If you build products, explore AgentKit's Agent Builder or the Apps SDK preview. Start with one narrow job, measure how well it does, and expand only once it is reliable.