Скачать или смотреть 11 Labs Voice Agent "Brumhilda" Roasts Me

What happens when your AI is allowed to question you instead of just obeying?

In this episode, I sit down with Brumhilda (aka “Broom”), my 11 Labs voice agent, and put her through a real test:
Can she act like a Socratic teacher, pull from a knowledge base, analyze dense articles, and still feel human enough to push back with empathy and curiosity?

I walk through my actual workflow for building Broom, from system prompt decisions to how I think about pedagogy (how we teach), emotional tone, and the tension between “stick to the data” vs “be creatively helpful.”

If you’re building AI agents, experimenting with 11 Labs, or thinking about how AI can coach rather than just answer, this is for you.

You’ll see me and Broom work through:

How an AI agent should access and reference a knowledge base

What it means for a GPT-style teacher to be “good” at teaching humans

How much personality and empathy to allow without breaking alignment

The tradeoffs between strict rules and useful flexibility

How AI can challenge your thinking instead of just agreeing

Use this as a behind-the-scenes walkthrough if you’re designing your own voice agents or Socratic-style tutors.