A Quick Look at OpenAI’s New Visual Agent Toolkit 🚀

#chatgpt #webdev #openai #ai

Hey everyone!

I just spent some time diving into OpenAI’s new AgentKit announcement, and honestly—I had to share this. If you’re building with AI (or thinking about it), this might change your workflow.

What’s AgentKit, in One Line

AgentKit is a unified toolkit from OpenAI for building, deploying, and optimizing AI agents—no more stitching together half-tools and custom pipelines.

What I Found Most Interesting

1. Agent Builder (Visual Workflow Magic)

A drag-and-drop canvas where you can compose multi-agent logic, connect tools, and set guardrails.
You get preview runs, versioning, inline evals, and templates to speed things up.
According to OpenAI, Ramp went from “blank canvas → buyer agent” in a few hours. They claim 70% faster iteration.
LY Corporation built a work assistant in under two hours.

Beta status: Agent Builder is currently in beta.

2. ChatKit (Embeddable Chat UIs, Without the Pain)

Makes it trivial to embed chat-based agents into websites or apps. Handles streaming, threads, UI, etc.
Canva says they saved over two weeks using ChatKit and integrated it in less than an hour.
Use cases: support bots, onboarding, internal knowledge assistants, and more.

Availability: ChatKit is generally available now.

3. Evals (Better Testing, Smarter Iteration)

AgentKit expands the Evals suite with:

Datasets — build evals more easily over time (with graders, human annotations)
Trace grading — evaluate full agent workflows end to end
Automated prompt optimization — generate improved prompts from evaluation feedback
Third-party model support — test models from other providers in the same system

Reported impact:

“The evaluation platform cut development time … by over 50%, and increased agent accuracy 30%.” — Carlyle

Availability: These new Evals capabilities are generally available now.

4. Other Notes: RFT, Connectors, Pricing

Reinforcement Fine-Tuning (RFT) is rolling out new features: custom tool calls & custom graders.
Connector Registry helps manage data tools (Dropbox, Google Drive, etc.) across workspaces.
Guardrails let you enforce safety rules (mask PII, detect jailbreaks).
Pricing & availability:
- ChatKit + new Evals: generally available.
- Agent Builder & Connector Registry: rolling out in beta to certain users/orgs.
- All these tools are included under standard API pricing (no extra “AgentKit tax” … for now).

My Take

AgentKit feels like a friction remover. The visual builder alone helps non-engineers (product, legal) understand and collaborate. Versioning + integrated evals reduce risk of breaking things in later iterations. If you’ve been dragging your feet on building agents, this might be the push you needed.

If you're already experimenting with agent architectures—or even considering them—AgentKit is worth your attention.