Fast and Hard Code

A prevailing sentiment on Twitter posits that "programming has been resolved." However, the reality is that familiarity with a programming language no longer holds significant weight, and some previously crucial human friction factors are now obsolete for automated agents. Consequently, Language Model (LLM) technologies render language selection considerably less vital than it once was.

If an individual is dissatisfied with their chosen language, they can seemingly translate it into another, allowing the LLM to opt for a language even the programmer might not be familiar with. This new paradigm has led to more programmers opting for languages based solely on their marketing appeal rather than technical merits. As a dedicated Rust programmer, I find it intriguing to observe the surge in Rust code being deployed by individuals who would have previously steered clear of it.

This trend can be attributed, at least partially, to two recent shifts: heightened emphasis on fast and efficient software, and the perception that LLMs are highly adept at optimizing code without compromising functionality. Renowned figures like Mitchell Hashimoto, Charlie Marsh, Jarred Sumner, and Daniel Lemire, who have long championed fast and performant software, are also receptive to agent-generated code.

This, coupled with other factors, has prompted others to join in. Many projects now prioritize speed and compactness, often favoring "hard" languages. It's not just Rust that is reaping benefits; Zig, despite the core community's reservations about AI, is also gaining traction. Notably, Cloudflare's Artifacts service utilizes a pure-Zig Git-protocol engine, while Vercel has introduced fx, a Zig coding agent marketed as small and fast.

These initiatives are predominantly driven by LLM assistance. More significantly, it's not just the adoption of less common languages but also the exploration of significantly more complex technologies. Developers are now creating impressive applications using advanced concepts such as DWARF files, Extended Berkeley Packet Filter (eBPF), custom network drivers, custom cryptography, and even vintage computing hardware.

Many of these domains were previously inaccessible to many developers. While this shift might lead to increased complexity, it also presents an opportunity for a broader range of developers who are eager to create fast and compact software.