Sep 3, 2026 · 1h 30m
AI agents face their hardest test in the handoffs between machines
The real danger is the gap BETWEEN two robots | E29
As AI moves into businesses, vehicles, laboratories, and robots, safety increasingly depends on how systems coordinate, authorize actions, and recover from mistakes.
- 1Agent safety requires authorization, observability, and deterministic controls—not just more capable models.
- 2Gatik’s autonomous trucking shows how local computation, real-world feedback, and regulation turn AI capability into operating infrastructure.
- 3Physical AI multiplies risk because every handoff between robots, tools, and machines creates new combinations to govern.
Don't miss
Gautam Narang explains how Gatik combines onboard computation, real-world deployment, synthetic data, and field feedback to improve driverless trucking.
The brief
Gautam Narang, Russ d’Sa, and Billy Kraft join Jason McCabe Calacanis to examine AI agents leaving chat windows for business workflows, vehicles, laboratories, and robots.
The panel argues that capability is not authorization: agents need shared context, observability, deterministic controls, and accountability before they can act broadly on behalf of people.
Gautam Narang describes Gatik’s driverless trucking operations, where onboard NVIDIA compute, local models, synthetic data, and field feedback support repeatable freight routes.
The episode’s central safety concern arrives at the system level: handoffs between devices create combinatorial risks that no single robot or model can fully explain.
A later debate over AI-writing detectors and watermarking broadens the question from machine safety to provenance, transparency, and the limits of measuring authenticity.
Books & mentions
Some links are affiliate links — PodLume may earn a commission if you buy.
