Why Rust Over Other High-Level Languages

For the past couple months I've been writing High-Level Rust - which gives 80% of the benefits with 20% of the pain.

One of the biggest pieces of feedback I've gotten is why not just use a high-level language that does this already? Think Go, C#, TypeScript, OCaml, etc.

Here's why I think, in the current model and language landscape, Rust is still the best choice.

Why Rust

• Expressive types + compiler hints - types scale logic to larger orgs and apps. I saw this transition first-hand at MasterCard, Facebook + Instagram, Rippling, and now Vanta. Expressive types allow you to precisely model the domain, making invalid states unrepresentable and removing whole classes of bugs. Rust's compiler goes further with ownership and lifetimes protecting against additional classes of bugs in data races and corruption - a category of protection no other mainstream high-level language offers.
• Mainstream long-term - Rust is now a default choice at the infrastructure layer. It's used in the Linux kernel, Android, AWS core, and Cloudflare edge. Rust proved itself at the infra layer and is now moving up the stack with more webapps and games built in it (I'm building CloudSeed Rust to make webapps easier). Rust has been the most admired language on Stack Overflow for 10 consecutive years with ~15% usage and ~20% YoY growth. The community continues to grow and these usages cement its longevity.
• AI proficiency - can one and few shot large features. In Why Rust Wins in the Age of AI, I cited benchmarks showing Rust winning not at speed or cost of AI generation but at real-world task completion - SWE-bench Multilingual showed Rust completing 58.14% of tasks, the best of 9 languages. Training data volume explains ~40-50% of cross-language AI performance variance and Rust's large, growing community means it has plenty. The large amount of training data combined with great types + compiler hints gives agents a tight feedback loop that leans into a pit of success.
• Great performance - High-Level Rust with liberal cloning takes a 10-30% perf hit but is still competitive with optimized Go and C#. With hot path optimization - switching clones to references and mutations - it often pulls ahead significantly.

Rust has historically gotten a bad rap for being too low level with steep learning curves and ongoing hits to developer velocity. But with AI agents doing most of the coding, you can now choose basically any language and minimize onboarding and velocity issues.

So why not Rust?

Common Objections

• "Just use Go" - Go is often perceived as the closest competitor with great perf, a big ecosystem, and lots of AI training data. But Go's type system can't make invalid states unrepresentable - it has no sum types or exhaustive matching. You get fast code that compiles but still lets entire categories of logic bugs through. Rust's expressive types close that gap.
• "Just use TypeScript" - Biggest ecosystem, most AI training data. But TS types are lies - they don't enforce at runtime (as evidenced by the prevalence of Zod), the ecosystem is in constant thrash, and the perf ceiling is low (not terrible, pretty good actually, but not great). TypeScript is a great language for high level apps but is less great for systems you want to be correct and function long-term - at least without constant maintenance.
• "Just use OCaml / F# / Gleam" - These have great type systems, some arguably better than Rust's. But AI is bad at them - FPEval (2026) found LLMs produce non-idiomatic imperative code even in functional langs. Training data volume explains ~40-50% of cross-language AI perf variance and these langs have a fraction of Rust's. Agents can help but they can't reliably finish full features autonomously in these languages, at least not yet. And it's unclear if this gap will shift - maybe with reasoning models to make up for the lack of data or maybe AI is bad at it and fewer people use it, leading to the programming language death spiral.
• "If AI removes language barriers, why not pick the easiest language?" - AI removes the floor but Rust's compiler raises the ceiling. Agents + Rust = fast AND correct. Agents + Python = fast but fragile. You're trading some agent speed for a lot of agent reliability which I think is a good tradeoff - move fast with stable infrastructure.
• "Rust's web ecosystem is immature" - Fair. Axum and actix are solid but young vs TS/JS's Express, dotnet's ASP.NET, or Go's stdlib. The ORM story is also weaker. This is a real trade-off but the ecosystem is growing fast and for most web apps the framework does only 20% of the work and your domain logic does 80% - that's where Rust's types shine.
• "Good luck hiring Rust developers" - Also fair, for now. But if AI agents are doing most of the coding, you need fewer deep Rust experts and more people who can direct agents and review output. Agents give matrix-like powers - I don't know this thing but I want to do it anyway. And you can, agents supercharge JIT learning and an agent trained on mainstream practices gets you average outputs at the outset. In many ways agents are raising the layer of abstraction to a point where the programming language is an implementation detail - you can focus on the system and its goals.
• "This could all change" - Totally. Models could get way better at niche languages, removing Rust's training data advantage. Languages could evolve to add new compiler-level checks like ownership and lifetimes - or entirely new constraints we haven't thought of - with simpler syntax. We may even get AI-native languages that are just far more effective than human-native ones. Those could leapfrog Rust, in which case I'll reconsider.

Next

Rust gives you expressive types, compiler-enforced correctness, mainstream longevity, strong AI proficiency, and great performance. AI agents remove the historical learning curve barrier, making Rust's ceiling benefits accessible without the floor costs.

So, for now, I'm choosing Rust.

If you liked this post you might also like:

• Why Rust Wins in the Age of AI
• High-Level Rust: Getting 80% of the Benefits with 20% of the Pain
• The Missing Programming Language - Why There's No S-Tier Language (Yet)