DevDay 2025: OpenAI launches agent kit, updates Codex model

OpenAI has introduced Agent Kit, a new set of tools on its platform designed to help developers and enterprises build artificial intelligence (AI) agents from prototype to production, CEO Sam Altman said on Monday.

Agent Kit includes a range of core capabilities, such as:

• Agent builder for creating and testing agent logic
• Chatkit for embedding customisable chat interfaces
• Evals for agents to measure performance and understand agent behaviour through trace grading.

Trace grading is the process to analyse and evaluate how an AI agent makes decisions step by step.


In addition to core capabilities, developers can check how different parts of an agent are working and use built-in tools to make their prompts smarter and more effective.

The kit is built on top of OpenAI’s Responses API, already used by hundreds of thousands of developers, Altman said on the sidelines of DevDay conference in San Francisco.

Additionally, it also integrates with OpenAI’s Connectors Registry, so agents can safely connect to company tools and other apps, through an admin dashboard.

Alongside Agent Kit, OpenAI announced new coding features powered by an updated version of Codex, now trained from the GPT-5 family of models. Per the latest updates, Codex can adjust its thinking time based on complexity of a task and is optimised for agentic coding workflows.

Since its release, Codex has already processed over 40 trillion tokens, highlighting its wide adoption among developers, Altman revealed.

The number of developers using OpenAI’s tools has also doubled to 4 million, and API usage has increased from 300 million tokens per minute to 6 billion, he said.

OpenAI also introduced apps that can run directly inside the chatbot. Apps including Booking.com, Canva, Coursera, Expedia, Figma, Spotify, and Zillow already made available, starting Monday.