AI Coding Weekly

Thorsten Ball: agents don't need unit tests

Thorsten Ball says more than half of human written unit tests become useless once the program works, and agents do not need the same ladders. Lachlan Donald says that is exactly backwards for large codebases.

The big disagreement

Thorsten Ball says agents do not need unit tests the way humans do

Thorsten Ball argues that TDD and unit tests exist because of human limits, short-term memory and distraction, and wagers over half of all human written unit tests become useless once the program works. Lachlan Donald replied that good unit tests matter more with agents, because they define behavior at the finest grain and keep the context loop small.

Lachlan Donald
@lox
X
Unit tests are behavior specifications that scale with codebase size.
Oct 4, 2026 · View on X

Read the full story

Receipts

Bun's ahead-of-time compilation starts Claude Code 24% faster

Jarred Sumner posted numbers from an early preview of Bun plus JSC with ahead-of-time compilation: 24% faster startup, 48% lower memory and 46% lower CPU for the Claude Code CLI. In a follow-up he says Claude Code is now 304 MB instead of 430 MB, and that TypeScript 6 runs faster in Bun AoT than in Bun JIT.

Jarred Sumner
@jarredsumner
X
Other workloads are different though, TypeScript 6 runs faster in Bun AoT than in Bun JIT now
Oct 4, 2026 · View on X

Read the full story

How they actually work

Built from 502 posts by the engineers, language designers and agent builders we follow on X over 1 days.

Get the next one by email

Coding with AI, read daily so you do not have to. The experiments, the receipts and the arguments from engineers shipping real software. Not a changelog.