On Monday, OpenAI CEO Sam Altman revealed AgentKit, a toolkit designed for creating and deploying AI agents, during the company’s Dev Day event.
“AgentKit offers a comprehensive suite of modular tools on the OpenAI platform, enabling you to move agents from initial concept to full deployment. It provides everything necessary to build, launch, and refine agent workflows with much less hassle,” Altman explained.
This release underscores OpenAI’s commitment to making agent development more accessible and efficient for developers. It also marks a strategic move to compete with other AI platforms that are striving to deliver all-in-one solutions for building enterprise-grade autonomous agents capable of handling sophisticated tasks beyond simple prompt responses.
AgentKit was among several new features introduced at OpenAI’s Dev Day, alongside the rollout of app-building capabilities directly within ChatGPT, which now boasts 800 million weekly active users.
AgentKit is equipped with several fundamental features. The first is Agent Builder, which Altman likened to Canva but for designing agents.
“It’s a quick and visual method for mapping out logic, processes, and concepts,” Altman stated. “It’s built on the responses API that hundreds of thousands of developers already rely on.”
The next feature, ChatKit, delivers a straightforward, embeddable chat interface, allowing developers to integrate chat functionalities into their own applications.
“You can incorporate your own branding, custom workflows, and any unique elements of your product,” Altman added.
Evals for Agents brings in tools for evaluating AI agent performance, such as detailed trace grading, datasets for testing individual agent modules, automated prompt tuning, and the capability to run assessments on external models directly from the OpenAI platform.
AgentKit also grants access to OpenAI’s connector registry, enabling developers to safely link agents to internal resources and third-party services via an “admin control panel,” all while preserving security and oversight.
To showcase AgentKit’s simplicity, OpenAI engineer Christina Huang constructed a complete AI workflow and two separate AI agents live on stage in less than eight minutes.
“These are the tools we always wished we had when we started building our first agents,” Altman remarked, adding that OpenAI has already partnered with several early adopters who have successfully scaled agents using AgentKit.
