OpenAI Unveils AgentKit and chatgpt Apps, Expanding AI Platform Capabilities
Table of Contents
OpenAI is doubling down on its platform strategy, announcing new tools designed to empower developers and bring more sophisticated, app-like experiences to users. The news, delivered during the company’s annual developer conference in San Francisco this week, follows recent launches like the Sora video creation app and signals a clear focus on building on top of OpenAI’s existing AI models.
The two central announcements are the beta release of AgentKit, a no-code agent builder, and the launch of apps within ChatGPT, aiming to transform the popular chatbot into a more versatile application platform.
building AI Agents wiht AgentKit
According to OpenAI co-founder and CEO Sam Altman, “AI has moved in the last couple of years from systems that you can ask anything to, to systems that you can ask to do anything for you – and we’re starting to see this through agents, software that can take on tasks with context tools and trust.” However, Altman acknowledged the current hurdle in agent development: it requires meaningful engineering expertise. agentkit aims to lower that barrier.
The tool allows developers to build agents without writing code, leveraging OpenAI’s models and a visual interface. AgentKit also incorporates safety features, allowing developers to set guardrails to prevent issues like jailbreaks and data leaks. It integrates with Evals, OpenAI’s tool for prompt testing and model behavior analysis.
To facilitate the deployment of these agents,OpenAI simultaneously launched ChatKit,a toolkit for embedding agent experiences directly into applications. Several OpenAI customers, including HubSpot, are already leveraging this service for both internal and external applications, such as customer support.
Apps Arrive in ChatGPT
The second major declaration centered on the launch of apps within ChatGPT – available to all users outside of the European Union. This feature allows users to invoke third-party applications directly within the ChatGPT interface. Developers can build these apps using the Apps SDK, which is built upon Anthropic’s MCP. The Apps SDK, being open source, enables developers to create both the logic and user interface of their applications.
Imagine a user seeking homes in a specific location. They could simply type “Zillow show me homes in Portland, OR under $500,000” into ChatGPT, and the Zillow app would display a map and interactive interface, allowing for refined searches and detailed property data.
The SDK also supports user authentication, enabling developers to offer personalized experiences and access to premium features. “The magic of this new generation of apps in ChatGPT is how they blend familiar interactive elements – like maps, playlists and presentations – with new ways of interacting through conversation,” the company explained in a release.
The initial app lineup includes integrations from Booking.com, Canva, Coursera, Expedia, Figma, Spotify, and Zillow. OpenAI plans to launch a public directory of apps adhering to its guidelines in the near future.
OpenAI acknowledged that data sharing controls within the SDK are still evolving and will soon offer more granular options for developers.A key challenge will be app revelation, mirroring the early struggles of voice assistants like Amazon’s Alexa and Google Home.While established services are likely to gain traction, newer developers may face visibility hurdles. OpenAI has indicated it will proactively suggest services, raising questions about how it will manage competition. The launch also comes with the shadow of GPTs, OpenAI’s previous attempt at ChatGPT apps, which has seen limited adoption.
Additional Updates: Codex, New Models, and GPT-5 Pro
Beyond AgentKit and ChatGPT apps, OpenAI announced the general availability of Codex, its AI coding agent, complete with a Slack integration and an SDK for embedding its capabilities into CI/CD pipelines. The SDK is currently available for TypeScript, with support for additional languages planned.
OpenAI also unveiled new, more cost-effective AI models: gpt-realtime-mini, a real-time voice model 70% cheaper than its larger counterparts, and gpt-image-1-mini, an image model 80% less expensive to use.
for developers seeking access to OpenAI’s most powerful models, GPT-5 Pro is now available through the API.
