This interview features John Janget, Principal Engineer at the AI defense startup Helsing and former Amazon engineer, discussing the trajectory, adoption challenges, and technical advantages of the Rust programming language. Janget, who holds a PhD from MIT, argues that Rust’s success stems from its ability to offer safety, performance, and ergonomics simultaneously, making it highly suitable for safety-critical (defense, automotive, embedded) and high-performance applications (like those requiring low tail latency at AWS).
Janget critiques the current state of AI, expressing skepticism that Large Language Models (LLMs) possess genuine understanding or reasoning abilities, calling the technology "overhyped." He suggests that AI will primarily accelerate repetitive coding tasks rather than replace skilled software engineers, particularly in complex domains like Rust due to its unique memory safety concepts (the borrow checker). He also details his career move from the high-salary, highly competitive US tech market back to Norway, driven by societal considerations like healthcare.
John Janget on Rust, AI Hype, and the Future of Software Engineering
0:36 Rust's Core Philosophy: Rust attempts to deliver the rare combination of safe, performant, and ergonomic development, traditionally difficult to achieve in a single language (e.g., ergonomic languages like Go often sacrifice speed for a garbage collector).
1:07 Background and Early Rust Adoption: John Janget, a Rust evangelist and live coder, completed his PhD at MIT, describing the environment as highly collaborative and curious. He began using Rust for his research database prototype after recognizing the limitations of his initial draft written in Go.
3:37 Barrier to Enterprise Adoption: Despite being the "most desired" language on Stack Overflow for nine years, Rust adoption remains low (12-14%). The primary reason is the extremely high switching cost for companies with established codebases and existing talent, rather than the language's learning curve.
7:25 Rust at AWS: Janget supported the internal Rust build system at AWS. Rust adoption was grassroots, driven by teams seeking performance benefits, especially reduced tail latency (a major struggle for runtime languages like Java/Kotlin due to garbage collection).
Go Teams: Rust makes runtime crashes (business logic errors) much less likely due to its strong type system and compile-time guarantees.
C++ Teams: Rust eliminates concurrency bugs statically at compile time (data races), a massive advantage over relying on dynamic analysis tools like Address Sanitizer.
13:06 Key Areas of Success: Rust is successfully implemented in embedded development (as a safer C/C++ alternative), command-line tools, and safety-critical applications (defense, automotive, space).
16:40 US vs. Europe (Norway): Janget moved from the US to Norway primarily due to dissatisfaction with US societal structures, particularly the approach to healthcare.
20:07 Salary Disparity: Compensation is significantly higher in the US. A senior Rust engineer in Norway might earn around $150k annually, whereas a similar role at a major US company (like Amazon) could reach $400,000 in total compensation, even after accounting for higher US expenses.
21:34 Rust in Defense/Safety: Janget acknowledges the community discomfort with his role at Helsing (AI defense startup) but emphasizes that Rust is essential because safety criticality is paramount. Rust's type system can encode complex invariants (e.g., preventing errors in spatial math transformations, 23:07).
31:04 AI is Overhyped: Janget is skeptical of LLMs, believing they are highly skilled at pattern replication but lack true understanding or reasoning (e.g., they can predict planetary movements but cannot deduce the underlying physics model).
36:46 AI’s Impact on Jobs: The threat of replacement is real for highly repetitive, pattern-based jobs (e.g., WordPress template design). However, for innovative and complex software engineering, AI will function as an accelerator tool, allowing developers to focus on higher-level design.
39:40 Rust and Generative AI: Generating correct Rust code is inherently harder for LLMs than other languages (like Python or Java) because it requires navigating the complex, unique rules of the borrow checker and lifetimes, a mental model the AI doesn't possess.
41:10 AI Infrastructure Strategy: Rust is generally inappropriate for machine learning research (where Python excels in rapid iteration), but it is the ideal language for production deployment of AI models in safety-critical environments (e.g., automotive).
51:49 The Borrow Checker: While often cited as the hardest part of learning Rust, the borrow checker is valuable because it forces developers to develop a deep understanding of data flow and lifetimes, making them better programmers even when writing code in other languages (Go, JavaScript, C++).
56:11 Best Compiler Feature: The Rust compiler's error messages are considered its best feature—they are concise, helpful, and instructive, aiding the developer in building the correct mental model of the language.
1:14:49 Corporate Influence: Janget asserts that the direction of Rust development is largely driven by its technical teams and community, and is not significantly steered by the mandates of large corporations.
1:18:16 The Uniqueness of Rust: If Rust disappeared, the world would lose the only language that successfully combines high-performance capabilities with modern safety and strong community-driven governance, leaving a vacuum that no current language (Zig, Swift, D) could fully fill.