What Is AI Agent Builder And How Does It Work? OpenAI Launches New Set Of Tools For Developers

At the heart of AgentKit is Agent Builder, a visual canvas where developers can drag, drop and configure their dream agent. No more endless code reviews or mystery logic. With clear drag-and-drop nodes, connecting tools, and configurable guardrails, building an agent feels less like wrangling code and more like constructing with digital LEGO. Agent Builder features full versioning, easy preview runs and inline evaluation settings, making iteration fast and frustration-free.

For those managing large organisations or complex workflows, the new Connector Registry arrives as the admin’s best mate. It brings every data source for example Dropbox, Google Drive, SharePoint, Microsoft Teams, and more, under one unified panel for easy access and security.

Added to this power-tool belt is Guardrails, an open-source safety layer. Guardrails mask personal data, flag risky behaviour, and generally keep agents from going rogue. The best bit is that it works alongside both Python and JavaScript out of the box.

ChatKit handles streaming replies, thread management, and the subtle art of showing just enough “AI thought process” to keep users engaged. With ChatKit, developers can now layer in chat UIs that blend perfectly into any site or app in minutes.

OpenAI said that developers can build robust agent evaluations using:

• Datasets: Rapidly build agent evals from scratch and expand them over time with automated graders and human annotations.

• Trace grading: Run end-to-end assessments of agentic workflows and automate grading to pinpoint shortcomings.

• Automated prompt optimisation: Generate improved prompts based on human annotations and grader outputs.

• Third-party model support: Evaluate models from other providers within the OpenAI Evals platform.

Developers can level up their agents with reinforcement fine-tuning. AgentKit is now available for OpenAI’s o4-mini model and in private beta for GPT-5, RFT lets teams teach their agents exactly when (and how) to use specific tools, and to evaluate themselves based on custom, use-case-defined rules.

There are two new beta features called the custom tool calls and custom graders which add even more control to that process.