LLMs let you choose the best language / stack for the problem you’re solving. My experience is that I’m able to build very performant Rust to solve some niche problems without having written a line of Rust code in my life. Without LLMs I would have defaulted to the languages I know and/or produce sub-par results trying to do in Rust.
I choose the best language / stack / framework for my use-case and then focus on building.
There is a lot of minutia and details in languages, frameworks, etc that I largely ignore whereas a few years ago I would actively try to convince someone to use some stack XYZ because of reasons.
"BigHits is a TypeScript 5 application using Svelte 5, SvelteKit 2 and Vite 5, with the Node.js adapter. It runs on Node.js 26 with npm 11.20.0 through Phusion Passenger and includes server-side API endpoints and a separate background TTS worker."
You can ask: "For hosting: can this app can be distributed across multiple 'boxes' with varying prices? How to optimize for ____?"
And then arrive at some sort of hybrid solution. The hosting environment is key - that informs of the possible stack choices.
So "know the canvas you are starting on, and then optimize on these as pre-requisites according to the intended direction" is the takeaway.
Isn't that what life's about or should be about? Or else answering, "how to take the non-mainstream, mainstream?"
If I need to do some systems level engineering, I’ll reach for Go or Rust. There is generally no reason to use Rust first, because 99% of us will not need 99 percentile performance. I also find Rust to be a tougher language. 98 percentile performance with Go is fine, where there are usually one or at most 3 ways to do something.
If I need web API services, I’ll use typescript for the ecosystem. Most providers offer SDKs and the broader ecosystem is rich. I can use node or bun, and these services can scale nicely at a reasonable $/maintenance/performance ratio.
I’ll reach for Docker because compute is now expensive and dedicated hosting is cheaper than AWS/GCP. I can get a 2020-2022 processor with 128gb RAM and generous bandwidth for $100-200 a month.
If you enjoyed Lisp, you should definitely check out Scheme, Racket, and Clojure. Scheme is my personal favourite language. OCaml is also great.
I use s7 Scheme hosted inside C apps. One of the cool things about Scheme is that are a lot of different ways of using it. (embedded, compiling to C, standalone, etc).
I always enjoyed clojure. It had a steeper learning curve than something like python, but you could build really powerful systems and abstractions. Now a lot of that joy is gone and you're probably best off going for the most performant language (ex. Rust, go etc)
I tried rust but the compilation times often make it hard to get into a flow state. I have no idea why it takes so long to compile. I thought low level languages compiled faster.
I still prefer vanilla TypeScript whether writing a browser application, command line application, or both. The language is extremely mainstream but the mere mention of not using common abstractions will trigger holy wars from people who cannot program.
LLMs let you choose the best language / stack for the problem you’re solving. My experience is that I’m able to build very performant Rust to solve some niche problems without having written a line of Rust code in my life. Without LLMs I would have defaulted to the languages I know and/or produce sub-par results trying to do in Rust.
I choose the best language / stack / framework for my use-case and then focus on building.
There is a lot of minutia and details in languages, frameworks, etc that I largely ignore whereas a few years ago I would actively try to convince someone to use some stack XYZ because of reasons.
Just built HN BigHits
"BigHits is a TypeScript 5 application using Svelte 5, SvelteKit 2 and Vite 5, with the Node.js adapter. It runs on Node.js 26 with npm 11.20.0 through Phusion Passenger and includes server-side API endpoints and a separate background TTS worker."
You can ask: "For hosting: can this app can be distributed across multiple 'boxes' with varying prices? How to optimize for ____?"
And then arrive at some sort of hybrid solution. The hosting environment is key - that informs of the possible stack choices.
So "know the canvas you are starting on, and then optimize on these as pre-requisites according to the intended direction" is the takeaway.
Isn't that what life's about or should be about? Or else answering, "how to take the non-mainstream, mainstream?"
With LLMs, I choose tooling based on the task.
If I need to do some systems level engineering, I’ll reach for Go or Rust. There is generally no reason to use Rust first, because 99% of us will not need 99 percentile performance. I also find Rust to be a tougher language. 98 percentile performance with Go is fine, where there are usually one or at most 3 ways to do something.
If I need web API services, I’ll use typescript for the ecosystem. Most providers offer SDKs and the broader ecosystem is rich. I can use node or bun, and these services can scale nicely at a reasonable $/maintenance/performance ratio.
I’ll reach for Docker because compute is now expensive and dedicated hosting is cheaper than AWS/GCP. I can get a 2020-2022 processor with 128gb RAM and generous bandwidth for $100-200 a month.
OxCaml is interesting you can choose the layout of the memory and finely tune the GC https://oxcaml.org/
If you enjoyed Lisp, you should definitely check out Scheme, Racket, and Clojure. Scheme is my personal favourite language. OCaml is also great.
I use s7 Scheme hosted inside C apps. One of the cool things about Scheme is that are a lot of different ways of using it. (embedded, compiling to C, standalone, etc).
I always enjoyed clojure. It had a steeper learning curve than something like python, but you could build really powerful systems and abstractions. Now a lot of that joy is gone and you're probably best off going for the most performant language (ex. Rust, go etc)
I tried rust but the compilation times often make it hard to get into a flow state. I have no idea why it takes so long to compile. I thought low level languages compiled faster.
I still prefer vanilla TypeScript whether writing a browser application, command line application, or both. The language is extremely mainstream but the mere mention of not using common abstractions will trigger holy wars from people who cannot program.