Programming 2026-10-08 3 videos

Building Reliable Software for Fast AI Change

Fast AI-driven change stays safe when systems isolate failures, control concurrency and verify behavior at multiple levels.

Summary

AI-era reliability requires architecture that contains overload and failure, with isolation, redundancy, sharding, back pressure and decoupled services protecting critical databases as demand surges. Concurrent operations also need explicit safeguards: idempotency prevents duplicate effects, while locks and design changes address races, livelocks and deadlocks. Testing must combine fast unit checks, integration coverage and a smaller set of end-to-end workflows because code and tests generated from the same mistaken assumption can pass while the system remains wrong. PlanetScale applies these principles with Vitess, Traffic Control, database branching and deploy reverts, while AI mainly increases the speed and volume of changes rather than defining correctness.

The videos

Move Fast and Don't Break Things: Scaling Databases for the AI Era — PlanetScale ↗ AI Engineer

AI-era databases stay reliable through isolation, redundancy, sharding, back pressure and decoupling, with PlanetScale using Vitess for MySQL, Neki for Postgres and Traffic Control to contain agent-driven surges that have pushed some daily active users up 2x, 3x or 100x.

Race Conditions on a Human Scale - Computerphile ↗ Computerphile

Race conditions arise when operations overlap, as two ATMs can each approve a withdrawal of the last £100, while idempotency, separate controls and locks prevent duplicate effects, livelocks and deadlocks.

How to Test Your Code (Even When AI Writes It) ↗ ByteByteGo

Reliable AI-written code needs the testing pyramid—many fast unit tests, fewer integration tests and a small set of end-to-end checks—because shared implementation and test assumptions can let broken checkout flows pass.