#663 – February 08, 2026
a simplified version to understand how it works internally
I made my own git
6 minutes by Tony Strømsnæs
Tony built a simplified version control system to understand how git works internally. Git stores files as SHA-1 hashes in a directory, creates tree objects to track file structures, and links commits together through hashing. Tony chose SHA-256 for hashing and zstd for compression instead of git's defaults. The implementation in Rust handles basic operations like tracking files, creating commits, and checking out previous versions by storing and retrieving compressed objects.
Build AI agents that respond with interactive UI instead of text. No workflows, no code.
sponsored by Thesys
Build AI agents that reason dynamically and respond with charts, cards, forms, slides, and reports without creating any workflows manually. Create data agents, website copilots, sales assistants, ecommerce agents and more in just 3 easy steps – connect your data, add instructions, and customize style. Publish and share with anyone or embed on your site. Build now for free.
The deletion test
3 minutes by Chad Fowler
Chad introduces the “deletion test” for software systems: imagine deleting all the code and ask what knowledge would survive. If that feels terrifying, it reveals missing specifications, tests, and definitions of correctness. Chad argues that code should be replaceable, while understanding, evaluation, and invariants must be explicit so systems remain safe even when code is regenerated.
Automatic programming
2 minutes by Salvatore Sanfilippo
Antirez introduces “automatic programming” as writing software with AI while staying fully involved in design, decisions, and understanding. He contrasts this with “vibe coding,” where AI works with little guidance. He says that quality software comes from human vision and intent, not just code generation. AI is a tool; the work and ownership remain human.
Finding and fixing Ghostty's largest memory leak
7 minutes by Mitchell Hashimoto
A few months ago, users started reporting that Ghostty was consuming absurd amounts of memory, with one user reporting 37 GB after 10 days of uptime. In this post Mitchell gives an overview of what caused the leak, she looks at some of Ghostty's internals, and shares some brief descriptions of how they tracked it down.
Software performance engineering
5 minutes by Rico Mariani
Rico outlines eight core ideas of performance engineering that apply across languages and systems. He points out that performance problems are usually systemic, not local, and depend on memory, cost models, and architecture more than clever code. The focus is on thinking clearly about how systems really behave, measuring wisely, and preventing slow performance decay over time.
And the most popular article from the last issue was: