Nick Shirobokov

I help devs test AI agents and workflows 10x faster | uv add appmerit | prev. TensorSense

[Nick Shirobokov @nicksherrow]

@swyx @METR_Evals Imo this is the most important coding eval atm. I'm honestly so tired explaining my fellow ai researchers why I don't merge their PRs despite they solve the issue

[Nick Shirobokov @nicksherrow]

@ibuildthecloud I once tried to build a CRDT engine for syncing Python arbitrary objects between processes and it was a big mistake

[Nick Shirobokov @nicksherrow]

@zebassembly @charliermarsh we all really need Ruff #283 man :'(

[Nick Shirobokov @nicksherrow]

@zebassembly This one works for small/medium length tasks

[Nick Shirobokov @nicksherrow]

@codercody Feature flag!

[Nick Shirobokov @nicksherrow]

@mil000 > It ends up pushing founders into predatory hostels and other bad housing situations

Is this why some angry founder pulled a gun on partners lol?

[Nick Shirobokov @nicksherrow]

@nicolomagnante @ycombinator @arseniycodes What a beautiful product! Wish you all the best guys

[Nick Shirobokov @nicksherrow]

@Kev96790724 @anirudhbv_ce @OpenAI @GeminiApp @sentra_app Holy shit, that sounds cool. Do you have any public code examples?

[Nick Shirobokov @nicksherrow]

@Nick_Davidov I guess that’s what happened:
- 4.6 is very creative, but tends to ignore ~10% of instructions in complex prompts, some patterns are overfit
- Anthropic tried to make it follow prompt instructions more literally
- turns out a lot of engineers like when Claude thinks for them

[Nick Shirobokov @nicksherrow]

@garrytan Coverage is ‘executed unique lines of code by the test suite / total non-test lines of code’. Your way of calculating coverage is incorrect

[Nick Shirobokov @nicksherrow]

@techpoto @zeeg What’s your favourite one?

[Nick Shirobokov @nicksherrow]

@Nick_Davidov From what I see, there’s a strong negative correlation between how much effort someone puts into managing their social graph and their actual social status

[Nick Shirobokov @nicksherrow]

I don’t have a strong opinion on RLS overall, but exposing databases to agents via tools and relying on RLS for security is definitely a terrible idea.

The best pattern I’ve seen is using hydrated copies with sync engines + isolated getters and setters

[David Cramer @zeeg]

RLS was a mistake and folks exposing that level of complexity to less technical users is asking for trouble.

It was a mistake in Firebase. It’s a mistake in Supabase. It will be a mistake in the next product too.

I personally - even knowing how to secure it - would never touch it. It’s the worst security footgun you can imagine. One small mistake and your data is available to the world.

[Nick Shirobokov @nicksherrow]

Can we stop pretending microservices are AI models?

[Y Combinator @ycombinator]

LLMs are great for human in the loop applications, but fail at deterministic developer tasks.

@interfaze_ai is a new AI model that outperforms general LLMs on high accuracy tasks like: OCR, Object Detection, Web scraping, Speech-to-text, Classification and more.

Congrats on the launch, @yoeven and @khurdula!

https://t.co/vWEQt4h5Kg

[Nick Shirobokov @nicksherrow]

@codercody in SF all kinks are welcome!

[Nick Shirobokov @nicksherrow]

@zeeg Have you been there?

https://t.co/XFGdnPr43e

[Nick Shirobokov @nicksherrow]

@Nick_Davidov That’s exactly how do most enterprise AI products look like under the hood (and the reason why they are so bad)

[Nick Shirobokov @nicksherrow]

@effectfully @lorecirstea sorry i was in a rush :(

[Nick Shirobokov @nicksherrow]

@lorecirstea @effectfully @effectfully you said "what if the company doesn't make enough money". I corrected you by saying the company doesn't need to make money to give equity to employees. This correction is valid regardless of all other things we discuss in this thread.