"Should We Actually Trust This Thing?"
28–32 min · March 2026
▶ Now Playing
This Episode
When should you actually trust AI output? Kyle, Kate, and Morgan dig into two stories that expose very different failure modes — and arrive at the same uncomfortable question.
Story 1
Amazon & AWS
A mandatory all-hands after AI coding tools caused real production outages. Engineering leadership vs. the AI evangelists — inside the meeting that wasn't supposed to exist.
Story 2
Tesla Full Self-Driving
Trust calibration at 70mph. When does over-reliance become negligence — and who's liable when it does?
The AI Trust Framework
The aviation parallel: autopilot created two failure modes — automation complacency and automation distrust. The fix wasn't "don't use autopilot." It was building protocols that define precisely when the human is flying and when the machine is.
For any AI-assisted decision, ask:
- What are the stakes of this decision?
- What is the AI's track record in this specific context?
- Do I have a protocol for when to trust it — and when to take the controls back?
- Has my organization built that protocol, or are we discovering it through incidents?
Sources
- CNBC — Amazon's internal meeting on AI-related outages
- Tom's Hardware — Amazon calls engineers to address AI tool issues
- Awesome Agents — Six-hour outage, senior approval requirement
- Fortune — Amazon's response and denial
- CNBC — Amazon layoffs and anti-bureaucracy AI push
- NHTSA — Full Self-Driving investigations
- Tesla — Vehicle Safety Report
- Waymo — Safety data
The Cast
Kyle
Host. Opinionated. Expect a history drop.
Kate
The correspondent. Tight, sourced, no spin.
Morgan
The heartbeat. "Well, why though?"
Full Transcript
442 lines
▶
Full Transcript
442 linesLoading transcript...