Openai has launched an agent builder that the company hopes to eliminate fragmented tools and make it easier for businesses to use Openai’s systems to create agents.
Announced at Openai’s Day in San Francisco, AgentKit now allows developers and businesses to build agents and add chat functionality in one place that could compete with platforms like Zapier.
By providing a more streamlined way to create agents, Openai goes even further and becomes a full-stack application provider.
“To date, architectural agents mean juggling fragmented tools. Conflex orchestration with versions, custom connectors, manual evaluation pipelines, quick adjustments, and weeks of front-end work before launch,” the company said in a blog post.
Agentkit includes:
-
Agent builder. This is a visual canvas that lets you see the multi-agent workflow of what and versions of what developers have created
-
The Connector Registry is a central area for administrators to manage connections between Openai products. The Global Management Console is a prerequisite for using this feature.
-
Chatkit allows users to integrate chat-based agents into the user interface.
Finally, Openai said it would build a standalone workflow API and add Agent Deployment Tabs to ChatGpt.
Openai has expanded its agent ratings, added features such as datasets with automated graders and annotations, and added trace grading to perform end-to-end evaluation of workflows, automated rapid optimization, and support for third-party agent measurement tools.
Developers have access to several features in AgentKit, but Openai is gradually deploying additional features such as Agent Builder. Currently, Agent Builder is available in beta, but Chatkit and the new evaluation feature are generally available. The Connector Registry has launched a beta rollout to several APIs and ChatGPT Enterprise and EDU users.
Openai said pricing for the AgentKit tool is included in the standard API model pricing.
Agent Builder
To be clear, many agents are built using OpenAI models. However, companies often access GPT-5 through other platforms to create their own agents. However, AgentKit brings the enterprise into the ecosystem, making it less likely that you will need to tap other platforms frequently.
The company, demonstrated during December, said its agent builders are best for quick iterations. It also provides developers with visibility into how the agent is working.
During the demo, Openai developers created an agent that would suggest panels to read and watch the DVDA agenda. It took her less than 8 minutes.
Other model providers have seen the importance of providing developer toolkits to build agents and tempting businesses to use more tools. Google announced the agent development kit in April, expanding its construction of multi-agent systems “with code under 100 lines.” Microsoft, which runs the popular agent framework Autogen, has announced that it will bring agent creation to one place with the new agent framework.
Openai customers such as Fintech Company Ramp said in a blog post that the team can build procurement agents in hours rather than hours.
“Agent builders have transformed what once took months of complex orchestration, custom code and manual optimization into just hours. The visual canvas keeps the product, legal and engineering on the same page, reduces iteration cycles by 70%, and acquires agents in two sprints rather than two quarters,” Lamp said.
AgentKit’s connector registry allows businesses to manage and maintain data across workspaces and integrate data sources into a single panel that exerts both ChatGPT and APIs. It has pre-built connectors to Dropbox, Google Drive, SharePoint and Microsoft teams. It also supports third-party MCP servers.
Another feature of Agent Builder is GuardRails, an open source safety layer that protects against personal identifiable information (PII), jailbreak, and leakage of unintended or malicious behavior.
Bringing more chat
Since most agent interactions involve chat, it makes sense for developers to simplify the process of setting up a chat interface and connecting with the agent they just built.
“Deploying a chat UI for agents is surprisingly complicated: handling streaming responses, managing threads, displaying model thinking and designing engaging chat experiences.
The company said Chatkit allows chat agents to be embedded in the platform and embed them in apps and websites.
However, some Openai competitors are beginning to think beyond chatbots and want to provide agent interaction that feels more seamless. Jules, Google’s asynchronous coding agent, introduces new features that allow users to interact with the agent via the command line interface, eliminating the need to open a chat window.
response
The response to AgentKit is primarily positive, and some developers note that while simplifying the building of agents, it doesn’t mean that everyone can build agents.
Some developers will display the agent kit as a complement to the pipeline, not as a Zapier Killer.
Zapier debuted in 2024 with a no-code tool for building AI agents and bots called Zapier Central.
