Post:
If you’re still shipping load‑bearing code in C, C++, Python, or vanilla JavaScript in 2025, you’re gambling with house money and calling it “experience.”
As systems scale, untyped or foot‑gun‑heavy languages don’t just get harder to work with—they hit a complexity cliff. Every new feature is another chance for a runtime type error or a memory bug to land in prod. Now layer LLM‑generated glue code on top of that. More code, more surface area, less anyone truly understands. In that world, “we’ll catch it in tests” is wishful thinking, not a strategy.
We don’t live in 1998 anymore. We have languages that:
- Make whole classes of bugs unrepresentable (Rust, TypeScript)
- Give you memory safety and concurrency sanity by default (Rust, Go)
- Provide static structure that both humans and LLMs can lean on as guardrails, not red tape
At this point, choosing C/C++ for safety‑critical paths, or dynamic languages for the core of a large system, isn’t just “old school.” It’s negligence with better marketing.
Use Rust, Go, or TypeScript for anything that actually matters. Use Python/JS at the edges, for scripts and prototypes.
For production, load‑bearing paths in 2025 and beyond, anything else is you saying, out loud:
“I’m okay with avoidable runtime failures and undefined behavior in my critical systems.”
Are you?
Comment:
Nonsense. If your code has reached the point of unmaintainable complexity, then blame the author, not the language.
I agree with the post. Setting up typescript takes an hour or two if you have no clue what you’re doing. In return you get the absence of (the equivalent of) null pointer exceptions.
I chuckle every time I find an NPE in the Java backend. Doesn’t happen to me. Can’t happen to me.
Sidenote, while I’m already gloating: Once the backend code had an error where they were comparing two different kinds of IDs (think, user ID and SSN), which gave wrong results. This error can’t happen to me either, because I type my IDs such that they are not comparable. A strong type system really is a godsend.
Modern versions of Node.js have native TypeScript support. For scripts, you can just write the script then run it. That's it. No build process needed. A beginner could just rely on type checking in their editor (I think VS Code has the TypeScript tooling installed by default?)
For web apps, just use something like Bun or Deno. Bun gives you practically all the tooling you'd need (JS runtime, TypeScript, package manager, test runner, bundler, and framework for building web apps) out-of-the-box. It doesn't have a formatter, but you can just use your editor's formatter.
NPEs in Java usually have 2 causes and they are easily preventable:
The problem is not really language specific because quite the same can also happen in any programming language, the symptoms/errors are just different.
Nulls are absolutely pervasive in Java and NPEs are not avoidable. At minimum, most of the ecosystem uses nulls, so most any library will have nulls as part of its interface. Null is an inhabitant of every type in Java (even
Optional, ironically). You cannot escape it. It's a fundamental flaw in the design of the language.Btw, you also can't escape it in Typescript, either, due to unsoundness of the type system and the fact that many types for libraries are bolted on to the original JS implementation and may possibly be inaccurate. But still, it's a lot less likely than Java.
NPEs are the reason why my team moved to Kotlin. Well, that and all the other myriad advantages Kotlin brings to the table.
One of the backend devs started using Kotlin. From time to time I need to read backend code, and I have to say it’s very easy to write an incomprehensible mess in Kotlin.
Kotlin isn't perfect and it gives the devs quite a lot of freedom. I would argue that if your Kotlin code is messy, that's on you - but it will still be significantly less prone to failures like NPEs. Unless you opt out of null safety by using the dreaded ?-Operator.
I like the "advantage" that using any Kotlin code in normal Java automatically forces you to use the kotlin-stdlib for Java 8 and adds a additional bloat of 10MB to your project...
Well, ideally you start new projects writing 100% Kotlin while only adding Kotlin code to older codebases you can't get rid of. Personally, I don't like mixing languages anyway and I would stay with Java in Java projects. One reason is the bloat argument you pointed out quite correctly.
Ohhhhhh noooooooooo, not my 10 mbs
10MB? Oh no! How is this going to fit on my 1TB drive?
Lombok had a bunch of great things that should've been part of the java language to begin with. They've slowly been folded in (so now you have to work out which @NotNull annotation you want) but the language does still improve.
True... but Kotlin makes Lombok quite unnecessary by having its concepts built in. It's also worth to point out that null safety is opt-in in Java and opt-out in Kotlin.
That's a good point. You can get away with that with a new language, but adding nullability as a non-default option would break existing code much like making things const by default in C++ would, I suppose.
I really like how I can turn everything into immutable
val's and represent different paths as expressions, it can IMO really reduce the complexity of a function and makes it easier to spot bugs. I've been migrating some code of a FOSS app to Kotlin and was able to shrink most classes by like 30% while making it easier to read. The only thing I dislike about it is the additional syntax for various things, I could do without having multiple ways to write a constructor.Yes, there are things about Kotlin I don't love either. But I still like how it was clearly developed having developer quality-of-life in mind.
Recently I've just been getting co-pilot to do it
I have an alias that calls the copilot CLI with a prompt that says "set up typescript"
Fuck this
That's like one of the few good uses for LLMs in development - setting up new projects, but AI haters will downvote you no matter what.
I’ve seen VSCode advertise this for setting up tests, but I haven’t been trusting enough to actually try it.
Does it work consistently?