Why performant code matters (but gets widely ignored), with Casey Muratori

Casey Muratori emphasizes that despite overwhelming evidence of performance's critical role in software efficiency and cost, the industry largely neglects it. He discusses how most software runs significantly slower than necessary—by 10 to 100 times—and why this neglect persists due to misconceptions, architectural decisions, and a historical focus on rapid feature delivery over optimization. Casey advocates for integrating performance considerations into early design phases and measuring the theoretical limits of hardware to better target optimizations.

A key theme Casey advocates is for developers to learn how to read assembly language to understand what their code actually does at the hardware level. He explains how this skill reveals why high-level languages like Python are much slower compared to C or Rust—because of vastly different assembly instructions. Casey stresses that understanding CPU internals, including data movement, instruction flow, and execution units, enables developers to make informed decisions about architecture and optimize effectively. He notes that assembly is simpler than many believe and essential for performance tuning.

Casey discusses the importance of architectural choices in software development, stressing that good design should account for future optimization. He criticizes dogmatic practices such as excessive modularization or virtual functions that hinder compiler optimizations. Instead, he advocates for a pragmatic approach: understanding the hardware, minimizing serial dependency chains, and structuring code to keep performance bottlenecks identifiable and fixable later. Casey highlights how architectural decisions—like avoiding unnecessary serialization—are vital for scalable, high-performance systems.

Casey reviews how game development has shifted from individual studios building their engines from scratch to leveraging commercial game engines like Unreal and Unity, making game creation more accessible. He explains that historically, issues like engine risk and gameplay uncertainty led studios to rapid prototyping and vertical slices, enabling them to refine core gameplay early. He notes that modern live service games extend development cycles and market saturation increases, making discoverability challenging. Casey also discusses how the industry’s emphasis on cutting-edge graphics and multiplayer experiences drives long development timelines, exemplified by titles like GTA 6.

The podcast explores how access to powerful engines and AI tools influences the industry. Casey points out that while engines like Unreal reduce entry barriers and boost creativity, they also contribute to market flooding, where outstanding games struggle to stand out without marketing. Regarding AI, Casey notes that it’s still early to assess its full impact. He observes that AI tools are primarily used as augmentation rather than automation, and that full AI-driven development may still be years away. Casey warns that AI fatigue is emerging among developers, especially those with less autonomy, who feel their skills are devalued and their motivation diminishes.

Casey emphasizes curiosity and a deep understanding of hardware—such as CPU architecture and assembly language—as marks of a great developer. He criticizes dogmatic coding practices, highlighting how certain habits like excessive small functions or virtual calls can degrade performance by preventing compiler optimizations. He advocates for skepticism of received wisdom, relying on measurable, empirical evidence. Good coders should aim for simplicity, transparency, and maintainability, enabling hardware-aware optimizations and future improvements.

While acknowledging AI’s rising presence, Casey prefers traditional handcrafting and deep knowledge, viewing AI as a tool to enhance, not replace, craftsmanship. He draws parallels to traditional craft industries, emphasizing that some aspects of software and hardware creation are fundamental human activities. Casey discusses the psychological impacts, including burn-out and loss of motivation, caused by AI fatigue and organizational pressures. He suggests that autonomy in work fosters better AI integration and creativity rather than control through mandates.