Dmitry Zakharov

Welcome Sury 🚀 Retweet is appreciated 🙏 dev.to/dzakh/welcome-…

@artalar @SpaceXAI I have the same experience with GPT-5.6 Sol

I'm disappointed with @SpaceXAI's Grok 4.5 =/ It could be close to GPT-5.5, but it regularly makes tiny (but valuable) mistakes that Opus never does.

Here's the spec link standardschema.dev/json-schema

Made it tree-shakable as well for those who don't need it.

Surprisingly underrated Standard JSON Schema spec is now supported by Sury. This is something every JSON Schema-driven tool should use now, in my opinion.

Also started experimenting with Valgrind to have a deterministic performance metric

My AI "harness" is growing: Instead of managing subagents, I decided to focus on a verification tool. Now going to use Zod to cross-check types, JSON Schema, and validation logic, plus a property-based testing foundation.

@artalar @rescriptlang @typescript In short words: The problem of a leaking abstraction.

@artalar @rescriptlang @typescript For product scenarios, ReScript generates very readable JavaScript, which is more performant than alternatives in TS. But for Sury, I want to squeeze every drop of juice. And it was very difficult to make agents write this kind of code while needing to check the generated output

I rewrote Sury from @rescriptlang to @typescript and don't feel good about it🫠

@artalar @rescriptlang @typescript But I think I'll be able to reduce bundle size even more and optimize in some cases after a few iterations.

@artalar @rescriptlang @typescript -2% bundle size, the same perf. It's very similar because I used unsafe ReScript to produce optimized low-level JavaScript. One nice win is that for a mixed TS and ReScript codebase, the library will be included once.

@saltyAom Probably the reason dev.to/dzakh/zod-v4-1…

Benchmark • 25 routes Elysia 2 AOT (none) startup time, and memory is almost similar to Hono Interestingly 1. Elysia + TypeBox is much faster than Hono + Zod 2. Elysia 2 can completely tree-shake TypeBox Hono + Zod is better than Elysia for serverless/edge no longer holds

Writing a test framework for Sury - strict validation and formatting - guiding errors - automatic golden values - property-based testing, TS type testing, ReScript, bundle size, and performance from one definition. This is how much I want AI to write Sury code for me 😅

Everyone should accept that it's not always true. And it will be less and less true with time.

hear me out: if you want to get any good at playing bass, you absolutely will get callouses on your fingers. they're a physical requirement to be able to play long and hard. and if you can't play for long, chances are you won't get very good. if ai plays all your bass lines, at best you're some form of composer, prompting the model to articulate notes the way you want. but if that's all you've ever done, chances are you won't be able to pick up a bass and play what you composed because you never even developed the callouses. so, following this parallel, what are the callouses of programming? what are the things that enable you to become a good programmer?

@simonw Rewriting Hasura Graphql Engine

What's the most ambitious project you have Fable working on right now?

Rewriting to Rust is fun. Yesterday I made @envio_indexer consume 2.5x less ram for one of our biggest client by rewriting a piece of logic to Rust.

Wrote about the largest experiment in Ezno: recording and replaying events kaleidawave.blog/posts/recordin…

@_chenglou @ssalbdivad @_chenglou Curious whether you still use Sury or migrated to something else? There was a big gap in new releases, so I can understand your decision.

Wrote up the recipe for how I made Sury's TypeScript type-checking 15.7× faster. Inspired by @_chenglou, and powered by Attest — an amazing tool from @ssalbdivad 🙏 dev.to/dzakh/making-m…