What Your AI Agents Really Are When You Strip Away The Hype
About this episode
Nicole Königstein is writing two O'Reilly definitive guides on transformers and AI agents while doing a PhD on self-improving AI. She proved expensive models burn 4x the tokens with worse results and delivered the line defining 2026's vibe coding debate: "You can't outsource the thinking."
Key moments
- 0:00 The woman writing the AI agents textbook — what's she like off stage?
- 2:37 How would you explain AI agents to a non-technical person?
- 9:51 Are AI agents just if-then rules with better prompts?
- 16:26 Her PhD is secretly about AI that improves itself
- 19:43 The thought experiment that kills "just automate everything
- 27:17 Building a patent filing system from scratch
- 33:23 Why adding more agents won't fix your system
- 38:47 Every additional layer burns tokens
- 47:27 Can hallucination ever be fully solved?
- 53:44 Why expensive AI models overthink and produce worse results
- 55:37 She built a tool that picks the right model
- 57:45 Skill files as contracts — governance without jargon
- 1:01:09 What happens when you vibe code production?
- 1:05:18 Will AI agents replace software engineers?
- 1:13:27 Ungoverned execution isn't recoverable
- 1:14:24 outro