Principal Engineer at Microsoft working on .NET runtime.
Warsaw, Poland
Joined November 2010
- Tweets2.4K
- Following182
- Followers5.7K
- Likes7K
Pinned Tweet
A single .md file with tips & tricks for high-performance .NET/C# code, mostly meant for AI, and mostly aimed at benchmarks against the crustaceans 🙃 github.com/EgorBo/BlazingFas… (no unsafe!)
Contributions welcome if anything's missing, cc @axeng200 @ben_a_adams
I've benchmarked many models for my "AI bug hunting" script. Here's the latest run with (incl. Gemini 4 Argon). My favorite combo so far is 6.1 Sol + Sonnet 5.5. Sonnet 5.5 seems to need at least xhigh reasoning to shine - but that puts it in Opus territory pricewise.
GPT + Opus (via Copilot CLI) helped me uncover 180 JIT bugs, each with a minimal C# repro using only safe code 😶 - most had been hiding from us for many releases github.com/dotnet/runtime/is…
(Deleted my prev. post JIC) .NET devs should really stop introducing blazing fast (de)serializers 🙂. The world is going through a period when anyone using e.g. GLM5.3 can easily find and exploit a memory safety issue - many projects are overwhelmed by 1000s of potential CVE reports, though not everyone chooses to report what they find. Deserializers often accept untrusted input from external sources, making them prime targets for attacks. Embrace safety over saving a few nanoseconds and fewer heap allocations.
Как перевести C# union на русский язык при помощи автоперевода learn.microsoft.com/ru-ru/do… 🤔
EgorBo retweeted
It's out! #dotnet
Performance Improvements in .NET 11 - .NET Blog 🚀
devblogs.microsoft.com/dotne…
Decided to check C# (it also emits a Trie for pattern matches like that) - it turned out it needed a little help on the Roslyn side for this case, filed github.com/dotnet/roslyn/pul… 🙃
On larger switches, both C# and Rust emit a Trie: godbolt.org/z/hcxWeMzjq
C# strings are utf16 vs utf8, so more instructions + for each case C# emits a fast path (is it the same string literal? = frozen object).
Unions in C# vs alternative. Hint: both suggestions have memory model issues making them unsafe (at least the 2nd one, 1st one is arguably too).
C# 15.0:
modern java (top)
rust (bottom)
the java you thought you knew is no more
ALT sealed interface Temperature { record Celsius(int val) implements Temperature {} record Fahrenheit(int val) implements Temperature {} } void main() { Temperature temperature = new Temperature.Celsius(35); switch (temperature) { case Temperature.Celsius(var t) when t > 30 -> IO.println(t + "C is above 30 Celsius"); case Temperature.Celsius(var t) -> IO.println(t + "C is equal to or below 30 Celsius"); case Temperature.Fahrenheit(var t) when t > 86 -> IO.println(t + "F is above 86 Fahrenheit"); case Temperature.Fahrenheit(var t) -> IO.println(t + "F is equal to or below 86 Fahrenheit"); } var number_str = "10"; switch (Result.of(() -> Integer.parseInt(number_str))) { case Result.Ok<?> ok -> IO.println(ok.val()); case Result.Err<?> err -> IO.println(err.err()); }; }
It feels like good open-weight models fall into 3 price tiers:
1) 24-64GB MacBook: Qwen3.8-27B Q4–Q8. Might work for simple tasks with a plan from a smarter model, but IMO just a toy.
2) 1-2× RTX PRO 6000: Qwen BF16 (big context, kv-cache) or DeepSeek-V4-Flash (with offloading). The latter might compete with Sonnet 4.6 (on paper).
3) $0.5M–$1M rigs like 8×B300: Kimi K3, GLM-5.2, or DeepSeek-V4-Pro. Great models, but probably worse economics than Codex/Claude Code subs. Unless you find 24x used A100s...
P.S. "Mom, look, I ran 3-bit Kimi K3 on 4 Macs!" demos are useless.
Let's find the best programming language for AI in 2026–2027. An ideal use case for AI is "fire-and-forget" (correct, safe, performant, universal, diagnosable) I crafted a prompt to evaluate top contenders from the TIOBE top 20 (excluding dynamic or obviously unsafe languages):
---
Compare Rust, TypeScript, Go, Swift, C#, Java, and Kotlin to determine the best language for AI in 2026–2027, and build a tier list.
My priorities (in order):
* Logic correctness without relying on 100% test coverage or runtime checks
* Code safety & CVE vulnerability potential
* Rich production diagnostics (enabling AI to quickly identify correctness and performance issues)
* Performance & RAM consumption
* Platform support matrix (mobile, desktop, web, embedded, etc.)
* Deployment complexity, payload/dependency size, and ease of deployment
* Cost per task (rework required for requirement changes, compiler time saved vs. running tests, etc.)
Research planned 2026–2027 features for each language regarding these criteria (e.g., safety, diagnostics).
Reply with a comparison table (`Language │ Correctness │ Safety │ Diagnostics │ Deployment │ Platforms │ Performance │ Cost-per-task │ Weighted`), followed by an explanation for each language.
Avoid bias from historical memory (don't use it), my personal preferences, or my job.
---
Try this in your favorite AI tool and share the results!
