Moving Off of Typescript, 2.5m Lines of Code

Main pain points with TS have become more obvious as the team grew and the codebase resulted in a multitude of different models representing the same thing.

- Prisma model representing the things (plural because Prisma generates a ton of variants for the different read/write scenarios) going in/out of the DB

- Zod models for OpenAPI generation

- Zod models for deserialization where we have `jsonb`

- DTO models at the boundary

- Additional front-end payload models that wrap the DTOs

A developer building a simple API endpoint do to a read will often up writing a handful of models for the same entity to move it back and forth...A lot of this work simply ends up being related to the loss of runtime types requiring a lot more modeling work creating a spaghetti of Zod and types.

Lots of papercuts in day-to-day and increasingly difficult to get otherwise competent engineers onboarded.

At least I'm not under the impression that this is a Silver Bullet. Will C# make it better? We'll see!

I think the big gains come from a more mature ecosystem of things that "just work". e.g. EFC vs Prisma or Drizzle with EFC having, for example, automatic change tracking and automatic up/down migrations.

The two NPM supply chain attacks this week also highlight another issue with the ecosystem in general.

We'll see how this change goes and evolves in Motion; C# is still relatively rare for startups, but at series C, it's no longer a seed stage startup and it increasingly feels like "it's just another distributed enterprise backend" (or at least it should be).

One sticking point is that the tools, while community and small projects, are all behind a license. But, they are free for small teams making under something like $1M/year in revenue.

Love C#. Performance is fantastic. The framework and tooling are very mature (with exceptions like hot reload ugh). Entity Framework is amazing, by far the best ORM I've ever used.

I don't feel like an expert by any means but I'm productive. Some days I do hit some stuff that takes me longer that I would've hoped but I've been able to solve everything I've faced.

I only use it for APIs though. The fullstack stuff is not great. Maybe if Microsoft improved hot-reload I would consider Razor with a Vite setup. Blazor is just useless for me. It's very cool from a technical standpoint but your frontend is 10x slower and more bloated than the worst JS solution you can think of.

I connect my JS frontend layer with OpenAPI[1]. It's been ok but the whole OpenAPI experience is not as smooth as I had hoped.

[1] https://openapi-ts.dev/

There is a positive aspect to this, as it makes decisions easier as well. For many areas there is often a single good choice, there aren't dozens of web frameworks you have to evaluate. There's ASP.NET Core and then nothing else for a long time. And for the most part those are pretty good, so it does not feel limiting.

And C# doesn't have to be too enterprisy and over-abstracted. The community does tend a bit too much towards unnecessary abstractions in my opinion, but you don't have to use it that way.

Absolutely no reason other than "C# icky" -- they ended up with a platform that is crazy fast (and fast scales way easier, it handles a dumb amount of traffic without a lot of crazy design)

Startup culture is toxic AF at times, bad engineering decisions for cargo cult stuff.

You have to be absolutely insane and/or highly inexperienced to even consider Javascript/Typescript on the back-end. Not only is the language a terrible match but the deep architectural failings of NPM have been clear for at least a decade.