OpenAI Launches AgentKit to Help Developers Build AI Agents

During the company’s Dev Day event on Monday, OpenAI CEO Sam Altman announced the release of AgentKit, a toolkit for creating and implementing AI agents which is a major surprise on process automation. 

The introduced AgentKit, according to Sam Altman is a comprehensive toolbox for creating production-ready AI agents, at DevDay 2025 on October 6. The centerpiece is Agent Builder, which Altman describes as “like Canva for building agents.” This visual drag-and-drop platform is a direct rival of Make, n8n, Zapier, and the whole no-code automation ecosystem.

A wide range of building blocks called AgentKit is accessible through the open AI platform and is intended to assist you in bringing agents from prototype to production. Altman stated that It provides all you need to create, implement, and streamline agent workflows with significantly less effort.

The launch demonstrates OpenAI’s efforts to boost developer adoption by simplifying and speeding up agent construction. It also represents an attempt to compete with rival AI platforms that are vying to provide integrated tools for creating enterprise-level autonomous agents that are capable of more than just responding to commands.

The ability to create apps directly within ChatGPT, which has 800 million weekly active users, was one of the announcements made at OpenAI’s Dev Day, along with AgentKit.  

AgentKit has some essential features. Agent Builder is the first, and Altman compared it to Canva for creating agents.

AgentBuilder is only one aspect of AgentKit. Evals for Agents (performance measurement), ChatKit (embeddable chat interfaces), Agent Builder (visual workflows), and Connector Registry (secure tool integrations) are its four main parts. AgentKit claims to transform your nodes, Zapier zaps, and bespoke API calls if you’ve been duct-taping them together.

Altman described it as a quick and visual method of designing the ideas, actions, and logic. “It is based on the responses API, which is already utilised by hundreds of thousands of developers.”

The second feature offered by AgentKit is ChatKit, which gives developers a straightforward embeddable chat interface to incorporate chat functionality into their own programs.

“Anything that makes your own product unique, including your own workflows and brand, can be brought,” Altman stated.

With Evals for Agents, you can test the performance of AI agents with capabilities like automated prompt optimisation, datasets for evaluating specific agent components, step-by-step trace grading, and the ability to evaluate external models straight from the OpenAI platform.

Lastly, through a “admin control panel,” developers may safely link agents to both internal and external tools using AgentKit’s access to OpenAI’s connector registry, ensuring security and control.

Christina Huang, an OpenAI engineer, demonstrated the ease of use of AgentKit by creating a complete AI workflow and hosting two AI agents live onstage in less than eight minutes. 

Altman stated that OpenAI has already partnered with other launch partners that have previously scaled agents using AgentKit, adding, “This is all the stuff that we wished we had when we were trying to build our first agents.”