Build Hour: AgentKit
- Source
- youtube.com
- Author
- OpenAI
- Date
Context
OpenAI's Build Hour on AgentKit explains what the toolkit is meant to replace. The presenters say building an meant writing orchestration in code that could break when versions changed, hand-coding secure tool connections, moving data into a separate system, tuning prompts by hand and building a chat interface from scratch. AgentKit, which they say launched at OpenAI's DevDay a couple of weeks earlier, bundles a visual workflow builder with versioning, an admin center called the connector registry, built-in evaluations with automated prompt optimization, and a customizable chat interface called ChatKit.
The demo builds the start of a sales assistant: a classifier agent is forced to output one fixed category (data, lead qualification or email), and the workflow then routes to a specialist agent. The data agent reaches Databricks through an server, a standard connector for tools. This is OpenAI's own walkthrough of a prototype, so it shows what the product does rather than how well the resulting agents perform.
- AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
- eval — A repeatable test for AI quality — a set of tasks plus scoring — used the way software teams use test suites, because model output is too variable to judge by eyeballing.
- MCP — The Model Context Protocol — an open standard that lets any AI assistant plug into any tool or data source without custom integration code.
Checking sign-in…
Loading comments…






