No no that's my experience also, what I do is more steps in the pipeline. Eg PRD -> User Story -> Taskwise plan -> Functional -> Unit Plus as much linting and type checking and codestyle as possible across the board

So following up on this, the answer is clearly NO. If you don’t clean their codestyle at all, they get confused much faster. I tried this with all 3 CLIs and also tried to watch carefully that every file they work with is relatively short (to control for context len effects).

Can absolutely confirm at HubSpot our productivity and cross-team knowledge raised by a huge margin with adoption of AI tools. We have deep integation and MCPs that have full access to our codebases, internal docs, engineering docs and relevant contexts from RFCs and PRs and whatnot integration to our codestyle, component system and infrastructure know how’s. The outcome is: - Quick PR creation/implementation by out AI Agent (Sidekick) for more simple tasks - Iterating on our FE and redesigning/making new components is extremely simpler and faster than before - Implementing something new and I get good reviews and instant refactors when I realize Im going off the standards adopted by teams - When I don’t know about something it easily teaches me about libraries and systems owned by different teams and how to integrate it. Does it make mistakes? Absolutely. Do I use it blindly? No. I’m the driver for the architectures, implementation of complex code and fixing the AI mistakes and hallucinations but it definitely creates a draft that most of the times it is usable. That indeed increased velocity absurdly.

Your example might possibly be improved (by providing various codestyle rules etc). Today it's additional work, but models are getting better i.e. you can see significant improvement year over year (not necessarily with every new model).

DeepThink, of course, not even close. Opus is good at following codestyle instructions, etc., I often write code with DeepThink, validate with GPT-5.1 Pro, iterate, then polish/beautify with Opus 4.5 Extended Thinking.

some thoughts on agentic coding: with gemini 3 & sonnet 4.5 it really has become good performance and mileage seems to depend on how much feedback the ai can get : lsp , interpreted vs compiled for example - ai seems to be really good with strongly typed languages like Go & Rust cause of gopls and rust-analyzer - gut feel llms are probably very bad at Ruby (lsps arent great metaprograming) really good at go specifically if you define your codebase and set rules and guidelines a codestyle doc in an md file llm output becomes relatively consistent planning / architecture phase is really important imo for stuff which cant be one-shotted -> eventually saves time hot take - codex & claude code are slow , i dont use them. Kilo Code & Zed have faster agentic coding lsp access run indexes which make for better contexting codex / claude code just keeps greping most of the time in large codebases

自分も基本的に全部.gitignoreしちゃいたい派ですね codeStyle とかは .editorconfig でいいと思うし、入れるなら codeInsightSettings.xml とかくらいかな 変な差分でやすいから、バージョン管理したいモチベーションができてから入れてく感じでいい気がします

$ nanoPrime init --website-development ⚡ Fast websites. 🔒 High security. 🎨 Beautiful design. 📈 Maximum conversions. We’re coding your brand’s next big move. Ready to deploy? #NanoPrimeZW #WebsiteDevelopment #CodeStyle #TechBranding #ZimbabweBusiness

I'm building a fork of Laravel react starter kit that has different opinions on tools and developer experience. Docker, Makefile, Bun as runtime, strict defaults and codestyle.

I don't know who needs to hear this but Consistency in your codebase is more important than using the 'perfect' pattern. Pick a style and stick with it. #CodeStyle

🧩 [TIP] – ¿Te gustaría que tus scripts en R sean más legibles, ordenados y fáciles de mantener? Con {ARTofR} podés aplicar buenas prácticas y comentarios estructurados desde el primer archivo. ✔️ Ideal para proyectos colaborativos y enseñanza ✔️ Estructura scripts con encabezados y secciones temáticas ✔️ Fomenta documentación clara sin perder agilidad 💡 Una guía sencilla para escribir código que otros (¡y vos misma!) puedan entender. ¡Tu yo del futuro lo va a agradecer! 🔗 github.com/Hzhang-ouce/ARTof… ✍️ Huanyuan Zhang-Zheng #RStats #Rtips #CodeStyle #ARTofR #CienciaDeDatos

Tabs or Spaces? Which one do you use to indent your code and why? Let's settle this age-old debate! 👇 #Programming #CodeStyle

🌐Code is a language, and every developer speaks with an accent. From indentation styles to commit messages, the way you write code tells a story. Are you deliberate, experimental, minimalist? What does your “accent” say about you? #SoftwareEngineering #CodeStyle #programmer

[Tip of the day] Write your R scripts like others will read them, because future you is someone else Comment generously, name clearly #rstats #CodeStyle

We sold our freedom to Google, so we could waste no more time arguing about codestyle and just do our work 😏

What’s the best way to name Flutter widgets? Logical names or wild creativity? #CodeStyle

Codestyle question: A or B?

🚀 C# Devs, which do you prefer? 🚀 ✅ var firstName = "John"; ✅ string firstName = "John"; var keeps things short, string improves readability. What’s your take? Reply below! 👇 Looking for a dev? DM me! #CSharp #DotNet #CodeStyle #SoftwareDevelopment #LookingForADeveloper

4. 思考の変更方法を「海馬スイッチ」としてオブジェクト化 5. 思考の終わりで再びトークン挿入 6. トークン挿入後に再びReflectプロセスをつなげる。 これでClaude 3.7 sonnet thinkは普段よりかなり長く考えてくれた。"CodeStyle"で考える指示が効いてる感じ。 会話履歴 claude.ai/share/584f2f6a-745…

#CODESTYLE #CODE $CODE @TheCode0X