The Startup Ideas Podcast I got a private lesson on OpenAI's NEW Agent Builder
337 snips
Oct 8, 2025 Discover how to harness OpenAI's new tools like Agent Builder, ChatKit, and Widgets for creating robust chatbot workflows. Learn to build a classifier that effectively routes user inquiries while enhancing customer support. The discussion highlights the ease of integration and the visual approach that allows non-technical users to design workflows. A live demo showcases the workflow's capabilities, and insights on leveraging these tools for business growth wrap up the conversation, offering practical advice for startups.
AI Snips
Chapters
Transcript
Episode notes
Demo: Lead Classifier Then Two Agents
- Amir demoed a workflow that classifies incoming messages as leads or existing customers and routes them accordingly.
- The lead path captures contact details and the support path queries a vectorized knowledge base to answer tickets.
Use Classifiers With Examples And Output Schemas
- Use a classifier node with example prompts to decide routing and include reasoning examples to justify classification.
- Configure downstream agents with specialized prompts, tools, and desired output formats (JSON or plain text).
Iterate Prompts And Tune Reasoning Levels
- Use ChatGPT or prompt enhancers to iterate agent prompts and 'enhance' them inside Agent Builder.
- Set reasoning level per agent: choose minimal for form-filling and higher reasoning for complex troubleshooting to save cost and improve results.