OpenAI has taken an additional step forward in its commitment to applied artificial intelligence, at a time when industry focus is shifting toward agents. The company has announced the launch of AgentKit, a comprehensive platform that aims to simplify the creation, deployment, and optimization of AI agents—the new intelligent assistants capable of reasoning, connecting to tools, and executing complex tasks.
As Sam Altman described it, it is “a complete set of foundational components available on the open AI platform, designed to help you take your agents from prototype to production. It is everything you need to create, implement, and optimize agent workflows with far less friction.”
As you might imagine, developing an AI agent is not a simple task.
Until now, developing an agent required combining multiple services and weeks of high-level work—demands that made it difficult to approach this process from scratch and, above all, increased development costs. With AgentKit, OpenAI seeks to reduce that process to just a few hours thanks to a set of tools that includes a visual workflow builder (Agent Builder), a centralized connector registry (Connector Registry), and a conversational interface kit (ChatKit).
If we were to dissect the new AgentKit, we could say that Agent Builder serves as the heart of the system: a visual canvas that allows you to drag and connect nodes, such as agents, conditions, or security filters, in order to define the system’s behavior. Although the idea is reminiscent of automation platforms like the popular open-source tool n8n, AgentKit is based on a different approach, focused on the reasoning and decision-making of the agent (the “intelligence”), rather than executing repetitive tasks.
“Agent Builder has transformed what previously required months of complex orchestration, custom code, and manual optimizations into just a couple of hours. The visual canvas keeps product, legal, and engineering teams aligned, dramatically reducing iteration cycles by 70%, and allowing an agent to go live in two sprints instead of two quarters,” explained a spokesperson for Ramp, one of the first companies to test the tool.
Surrounding the Agent Builder is the Connector Registry, which allows administrators to control how agents access business data through integrations with services such as Google Drive, Microsoft Teams, or SharePoint. Finally, ChatKit provides a rapid way to incorporate these agents into digital products: integrated chats that can be customized for customer support, internal assistance, or research purposes.
Alongside the launch, OpenAI has expanded its evaluation system (Evals), which now enables the creation of test datasets, automation of performance analysis, and automatic prompt optimization. This layer of measurement is intended to ensure that agents not only function properly but can also be systematically evaluated and improved, just as with any other software.
The company is also expanding its Reinforcement Fine-Tuning (RFT) program, which allows for the adjustment of reasoning models for specific tasks or criteria. According to OpenAI, this will enable agents to learn to use tools more efficiently or adapt to a company’s internal policies.
During the presentation, OpenAI stated that ChatKit and the new Evals functionalities are already available for all developers. Meanwhile, Agent Builder and Connector Registry are in the beta phase, available to API and ChatGPT Enterprise clients. The entire toolset is offered under OpenAI’s standard model usage pricing.
With AgentKit, OpenAI solidifies its commitment to a new generation of autonomous, connected, and customizable agents that could become the foundation for future intelligent applications.
Images: OpenAI
Your email address will not be published. Required fields are marked *
Δ