
AI Toolz Podcast - 2026-10-04 : agents fail 87% of the time
Microsoft and Hugging Face release ThinkingBox, a benchmark showing top AI agents fail to consistently complete real business tasks despite confident output. Plus, OpenAI's former safety lead calls the company's approach 'broken' and argues AI labs should operate like nuclear plants. A reality check on AI reliability in production.
The skinny
The skinny isn't ready yet — notes appear once the transcript is processed.