OpenAI rolls out AgentKit to make it easier for developers to build AI agents

OpenAI gives developers a faster way to build AI agents

OpenAI is making it easier than ever to build custom AI agents. At its 2025 Dev Day event, CEO Sam Altman announced AgentKit, a comprehensive toolkit designed to help developers transition from prototype to production with reduced friction.

The goal is straightforward: to transform the complex process of agent creation into a visual, efficient, and scalable process.

With AgentKit, OpenAI aims to transform how teams design, deploy, and optimize agents across various applications, websites, and business workflows.

AgentKit brings all the building blocks together

Altman described AgentKit as “everything you need to build, deploy, and optimize agent workflows with way less friction.”

Instead of juggling multiple tools, developers can now manage logic, chat interfaces, evaluations, and connectors inside one platform.

This all-in-one approach makes AgentKit more than just a development kit; it’s a foundation for enterprise-grade AI orchestration that can handle reasoning, automation, and human interaction at scale.

Agent Builder makes creating agents as easy as designing in Canva

At the heart of AgentKit is Agent Builder, a drag-and-drop visual canvas that lets developers build multi-step, multi-agent workflows in minutes.

Think of it as Canva for AI logic, where you can connect tools, define decisions, and preview how your agent thinks.

It includes versioning, guardrails, and inline testing, allowing teams to iterate quickly. Ramp, a financial tech company, said Agent Builder cut their agent development time from months to just hours.

ChatKit helps developers embed chat UIs instantly

OpenAI also introduced ChatKit, a customizable chat interface toolkit that helps developers integrate conversational agents into their apps or websites.

It handles complex tasks like streaming responses, managing threads, and showing the model’s “thinking” process, all with a few lines of code.

Companies like Canva and HubSpot have already utilized ChatKit to develop in-app assistants for support and onboarding, transforming traditional help pages into dynamic, AI-powered chat experiences.

Evals for Agents takes performance testing to the next level

Building powerful agents is one thing; measuring them is another. OpenAI’s expanded Evals tools enable developers to track agent performance through datasets, trace grading, and automated prompt optimization.

You can even evaluate third-party models directly in OpenAI’s environment. These features enable teams to refine their prompts, enhance accuracy, and benchmark reliability before deploying agents in real-world scenarios.

Early testers, such as Bain and Carlyle, reported up to 30% performance gains using Evals.

Connector Registry unifies tools under one secure hub

Enterprises can now manage all their data sources and APIs from a single location using Connector Registry. This centralized admin panel enables teams to securely connect OpenAI agents to services such as Dropbox, SharePoint, Microsoft Teams, and Google Drive.

It ensures that enterprise agents have access to the right tools without exposing sensitive data. The Registry also integrates with OpenAI’s Global Admin Console, allowing IT leaders to manage access and permissions across large organizations.

Guardrails add an essential layer of safety for agents

To ensure reliability, OpenAI built Guardrails directly into AgentKit. This open-source safety layer protects agents from leaking personal data, falling for jailbreak prompts, or behaving unpredictably.

Developers can configure guardrails visually within Agent Builder or deploy them as Python or JavaScript modules.

These safeguards are critical for enterprise adoption, ensuring that custom-built agents operate within approved parameters while remaining flexible and adaptable.

AgentKit simplifies what used to take weeks of setup

Before AgentKit, developers had to piece together orchestration layers, connectors, evaluation scripts, and UI frameworks manually. Now, OpenAI bundles all of that into one streamlined toolkit.

Developers can visually design agent flows, embed them with ChatKit, connect to real data, and test performance all in the same environment.

This drastically shortens the development cycle and eliminates much of the friction that kept AI pilots from scaling into production-ready systems.

OpenAI shows AgentKit in action with live demos

During the Dev Day keynote, OpenAI engineer Christina Huang demonstrated AgentKit’s power by building two AI agents live onstage in under eight minutes.

She connected logic nodes, applied guardrails, and deployed a functioning workflow, all without touching raw code.

The demo impressed attendees and effectively showcased the intuitive nature of the new visual. What once required an entire engineering sprint can now be prototyped in minutes.

Ramp and LY Corporation praise faster development cycles

Early adopters, such as Ramp and LY Corporation, shared real-world examples of how AgentKit speeds up production. Ramp said it slashed iteration cycles by 70% and brought an agent to life in two weeks instead of two quarters.

LY Corporation built a multi-agent workflow in under two hours, noting that the visual interface brought engineers and domain experts together on the same page. This rare collaboration made development smoother and faster.

AgentKit turns OpenAI into a full-stack platform

With AgentKit, OpenAI is no longer just a model provider; it’s evolving into a full-stack development ecosystem.

By integrating workflow design, UI, safety, and analytics under one roof, the company positions itself alongside major automation players, such as Zapier, Retool, and n8n.

The difference is that OpenAI’s agents aren’t just rule-based; they reason, learn, and adapt dynamically, powered by GPT-5 and other large language models that handle logic and language at an enterprise scale.

Developers can now iterate on agents in real time

AgentKit’s real strength lies in how quickly developers can iterate. The drag-and-drop canvas supports preview runs and version control, allowing teams to tweak workflows instantly and re-deploy with confidence.

Combined with inline evaluations, developers can pinpoint where an agent underperforms and fix it on the spot.

This real-time feedback loop eliminates guesswork, enabling teams to build more intelligent and reliable agents faster than ever before.

ChatKit brings conversational AI to any product or service

ChatKit makes adding chat-based agents to any digital product as simple as embedding a widget. Developers can customize their appearance, integrate branding, and even connect to proprietary databases for contextual responses.

It’s already being used by brands like Canva, HubSpot, and LegalOn to power customer support, internal knowledge tools, and interactive documentation.

With ChatKit, conversational AI becomes an integral part of the user experience, rather than a disconnected chatbot.

Evals expand into deeper optimization and grading

OpenAI didn’t just stop at evaluation; it added automation. With new Evals tools, developers can automatically grade agent responses, identify weaknesses, and generate optimized prompts.

The system even supports multiple models, allowing teams to compare GPT-based agents against external AI systems.

Over time, these feedback loops create self-improving agents that perform better with every iteration. It’s a quiet but powerful shift toward autonomous optimization at scale.

Reinforcement fine tuning pushes customization further

AgentKit now supports Reinforcement Fine-Tuning (RFT) for GPT-5 in private beta. This enables developers to train agents to make more informed tool calls and adapt to unique business objectives.

By combining RFT with automated evaluation, teams can fine-tune both reasoning and performance. The result is agents that learn more quickly, follow domain-specific rules, and produce more accurate outcomes, all while remaining aligned with the company’s ethics and compliance policies.

OpenAI isn’t slowing down, and neither are its partners. See what’s driving the surge in Broadcom soars after partnership with OpenAI to build new AI chip.

AgentKit lays the foundation for the age of agentic software

OpenAI sees AgentKit as more than a toolkit; it’s the foundation for agentic software, where applications think, plan, and collaborate autonomously.

As more companies adopt AI agents for sales, support, and research, AgentKit provides the scaffolding for building those systems responsibly and efficiently.

From startups to global enterprises, it signals a new chapter where creating intelligent, goal-driven agents is no longer a futuristic idea, but it’s a practical, everyday reality.

Even the smartest AI has its blind spots. Discover what an insider has just exposed: A former OpenAI researcher reveals ChatGPT’s errors.

What do you think about OpenAI’s new AgentKit, which allows you to make AI agents more easily? Please share your thoughts in the comments.

Read More From This Brand:

• OpenAI just moved into chipmaking
• OpenAI hires Apple experts for debut hardware product
• OpenAI reorganizes group responsible for ChatGPT’s persona design

Don’t forget to follow us for more exclusive content on MSN.

If you liked this story, you’ll LOVE our FREE emails. Join today and be the first to get stories like this one.

This slideshow was made with AI assistance and human editing.