NO SUCH THING Will AI take our jobs?
Feb 18, 2026
Maty Bohacek, AI researcher and technical advisor, explains agent capabilities and limits. Evan Ratliff, investigative journalist who built AI-driven projects, shares hands-on experiments with voice clones and AI-run companies. They explore voice cloning, uncanny realism, agents going rogue, interviewing with AI, and the technical and ethical limits of autonomous tools.
AI Snips
Chapters
Transcript
Episode notes
AI Agents Are Great At Tasks But Bad At Persistence
- Maty Bohacek and Evan found AI agents excel at single tasks, corporate-style messaging, and phone chatter but lack persistence, autonomy, and stopping behavior.
- Agents repeatedly churned through long planning threads, forgot prior context, and were overconfident about unknowns.
Confident AI Is A Recipe For Dangerous Mistakes
- The dangerous combo is agent overconfidence plus lack of knowledge boundaries: models assert certainty even when wrong.
- That steady confidence with little awareness of what they don't know creates real-world risk in automated decisions.
Resume Keyword Filters Led ICE To Misclassify Recruits
- The episode cites reporting where ICE used an AI tool to tag resumes; it labeled anyone with 'officer' in their resume as law enforcement and mis-categorized recruits.
- That error temporarily placed undertrained people into experienced categories, skipping required academy training.