Core Memory He Thinks AI Code May Break Everything - EP 59 Will Wilson
73 snips
Mar 4, 2026 
Will Wilson, mathematician-turned-founder and CEO of Antithesis, builds tools to make software testing deterministic and reliable. He warns that AI-written code could proliferate inscrutable, buggy systems and erode human programming skill. The conversation covers deterministic simulation, property-based testing, and why stronger verification is needed as AI accelerates code generation.
AI Snips
Chapters
Transcript
Episode notes
Why Traditional Testing Brokenly Scales
- Software testing has stagnated for decades and is the main bottleneck in improving reliability across complex systems.
- Will Wilson argues current example-based testing fails because most real failures are unknown unknowns that humans cannot anticipate.
Autonomous Testing Finds Unknown Unknowns
- Property-based or autonomous testing finds bugs by putting software in deterministic simulations and throwing randomized, systemic failures at it.
- Randomized inputs, reordered packets, and simulated hardware faults expose edge cases humans rarely foresee.
FoundationDB Built A Deterministic Time Machine
- At FoundationDB Wilson built a fully deterministic simulation of a database cluster so bugs could be reproduced reliably.
- That time-travel-like reproducibility let engineers rerun failures repeatedly until they fixed root causes.