Anthropic engineers finally showed how they actually use Claude Code internally 31 minutes of internal workflow that most Claude users will never see on their own here's what they cover: > how to set up project context files the right way > custom commands that save hours of repeated... work > hooks that make Claude behave exactly how you need > subagents and how to actually spec them properly "your agent isn't the problem, your spec is" the people who understand how Claude Code actually works inside Anthropic are shipping things everyone else thinks requires a whole team that's exactly why I put together a breakdown of Claude features most people have never discovered you can find it belowshow more

The project context file setup alone is a game changer. Most people treat Claude Code like a chatbot the ones winning are using it like a junior dev with full codebase memory.

The piece most teams skip is a proper repo level CLAUDE.md. I went from fighting prompt drift every session to letting context handle itself once I wrote a decent one. Setup cost pays back in a day.

@AnatoliKopadze i swear those custom commands are life savers. didn't know we could save that much time!

amazing watch Toly, thanks for sharing !

Best part is what they skip. No prompt gymnastics, no chain-of-thought scaffolding. Constraint files and clear task boundaries. Took us months of overengineering to land on the same approach.

finally we'll know how to use it right

“Your agent isn’t the problem. Your spec is.” That might be the most important AI engineering lesson of 2026. Bad input → bad output. Clear thinking → insane leverage. The best AI users aren’t better typers. They’re better system designers.

waiting for your reply ser

the problem was actually in the specification

thanks for dropping this alpha with us anthropic been cooking content lately

my feed is heaven thanks for this share Anatoli, very interesting

@Fink0x Appreciated!

"your agent isn't the problem, your spec is" is probably the most important takeaway here

Crazy how the gap between developers is slowly becoming: people who know how to prompt AI vs people who know how to structure systems for AI 👀

クロードコードの機能の解説楽しみにしています。

@LilysAI_ 要約して

Those custom commands sound like a game changer!

been wanting to see their internal setup for ages that your agent isn't the problem your spec is line is so accurate most people just throw vague prompts at it and wonder why the output sucks

クロードコードを実際にどのように使用するかをanthropicエンジニアがついに示しました。 内部のワークフローの31分は、ほとんどのクロードユーザーが自分で見ることはありません。設定方法に関するプロジェクトコンテキストファイル >繰り返し作業を数時間節約するカスタムコマンド >あなたが必要とするまさにクロードを振る舞わせるフック >サブエージェントとその実際の仕様方法 "あなたのエージェントが問題ではない、あなたの仕様が問題だ" クロードコードがAnthropic内部で実際にどのように機能するかを理解している人々は、他の人々が完全なチームが必要と考えているものを出荷しています。それが私がほとんどの人々が発見したことのないクロードの機能の解説をまとめた理由です。

Now claude can make google to replace google totally. 😂

yess abolsutely claude is really good at all these 17 features...today will be using prompt cachig rest others i have explored and got great outcomes

I like it a lot. seeing that I was 95% working like them is also validating. I just can figure out how they did the verfication dashboard. I don't think I can create one as I work on mobile apps, but if you have ideas or tricks let me know.

@grok summarize this video

Context files alone changed how I structure every project now. The custom commands section is the part most people skip but probably shouldn't.

your spec is the agent is a pretty useful way to think about it

the spec is the whole game huh i've been wiring mine for weeks now this hits different can't unsee it

That quote is the whole talk: your agent isn’t the problem, your spec is. Hooks and subagents do not fix a weak spec, they multiply it. Parallelize a vague spec and you just get wrong answers faster, in more files. Spec first, then scale. The rest is amplification.

The role of the engineer is shifting. You are no longer just writing code. You are architecting the process for the AI to execute.

spec quality is underrated. most agent failures look like model failures, but they are really missing context and weak operating boundaries.

Is it even possible to reveal such secrets without facing consequences?

Ugh, these dirty fucking Indians.

wtf internal claude specs are pure alpha but which hooks

“Your agent isn't the problem, your spec is” is probably the most important AI lesson right now. Context, commands, and structure matter more than hype. Seeing similar patterns discussed across and lately.

31 minutes of internal workflow most Claude users will never find on their own, this is the one to actually watch

wtf internal claude specs are pure alpha but which hooks

Agentic AI Is Here. The Question Nobody’s Asking: Can You Prove What It Did? #KnowYourAgenticAi #rumedominic