How to name things

How to name things
11 minutes by Koleman Nix

Koleman says that good naming is one of the most important skills in software, because vague language usually means vague thinking. Names carry meaning for teammates, future readers, and AI tools alike, so choosing them well is an act of empathy. Rather than following rigid rules, good naming means studying the essence of a thing, looking at how nearby concepts relate, and asking how a reader will understand the name in context. A bad name often signals a deeper design problem worth fixing.

Revel is hiring. Build the next generation of software for hardware.
sponsored by Revel

As an engineer at Revel, your code doesn’t just live in a browser. It helps the world’s most ambitious engineering teams test rocket engines, nuclear systems, hypersonic aircraft, and other breakthrough hardware. Built by engineers from SpaceX, Anduril, and Palantir, Revel helps teams test up to 5x faster and develop higher-performing hardware with greater safety and reliability. Apply here.

A faster way to convert a timestamp
7 minutes by Ben Joffe

Most datetime libraries compute hours, minutes, and seconds using a chain of division steps where each result depends on the previous one, making parallelisation impossible. By reordering the calculations to break this dependency chain, and using fixed-point multiplication tricks instead of division, you can cut the process from around 16 CPU steps down to just two multiplications.

Good APIs age slowly
6 minutes by Yusuf Aytas

APIs that look elegant at first often cause the most trouble later. The real test is how an API holds up after different teams use it, requirements shift, and people rely on behavior that was never meant to be a promise. Yusuf points out that most problems come down to boundaries: once something is visible, people build on it, whether you intended that or not. His advice: Expose as little as possible, because adding later is easy, but removing is painful.

A coding agent is six functions in a trenchcoat
8 minutes by Hadley Wickham

Coding agents like Claude Code and Cursor work by giving an AI model tools to read, write, search, and edit files, plus run shell commands. You can build a basic version with just three functions in around 30 lines of code. Adding targeted file editing makes the agent faster and cheaper, since the model only rewrites changed lines rather than entire files. Safety matters too, as path checks and scoped tools prevent the agent from accessing files outside the project.

How well do agents use test and verification techniques?
about 1 hour by Dan Luu

AI coding agents mostly fail to use testing tools and techniques effectively, even when explicitly told to. Telling agents to use property-based testing, formal methods, or TDD changes their surface behavior but rarely changes the underlying quality of tests. Default instructions with no testing guidance outperformed most specific approaches, because specific instructions mostly just caused agents to do useless work inside unfamiliar frameworks. Simple, targeted nudges away from bad defaults worked better than detailed skill documents or named techniques.

How Are the Best Engineering Teams Shipping Faster With AI?
sponsored by Cloudways by DigitalOcean

Specific decisions about their stack, deployment workflows, and infrastructure that most developers haven't caught up to yet. Prepathon'26 (free, online, Sep 22-23) is two days of sessions where those engineers share exactly what they changed and why. Modern stacks, AI-assisted workflows, production infrastructure. From the people doing it live. Seats are limited.

And the most popular article from the last issue was:

• The startup's Postgres survival guide