Loading video...
Video Failed to Load
Boris Cherny, the creator of Claude Code at Anthropic, just explained why most people aren't getting real results from Claude in this talk he breaks down exactly how most people skipped the setup entirely: - 90% of code at Anthropic is written by Claude itself - the 14% you... show more
185,708 views • 4 months ago •via X (Twitter)
41 Comments
Most people use Claude like ChatGPT… the top 1% use it like an operating system.
This is very interesting bro! Just bookmarked it
Thank youuuu a lot Videl
@eng_khairallah1 whoa, 90% is wild. makes me wonder how accurate it is without human tweaks though.
as always thanks for alpha bookmarked
glad to hear that habibi
For me the interesting part is not that everyone become engineer. It is that software building is moving outside only engineering team. Then the hard part is not only code anymore, but who owns it, how it is tested, what data it can access, and who maintains it later.
Most people don't get results because they hand the model a blank slate every time. The fix isn't a smarter prompt, it's context that persists, what you decided last week, what you already rejected. Memory is the feature. Prompting is the workaround.
🫡
🙌
This is awesome
thank youu my man
Amazing
thank youuu man
i mean yeah. even my entire product is built by AI :)
same here bro! i built everything using AI
great article mate, very useful for me. bookmarked. 🤝
Context is everything with Agents. The more you invest in setup, the less you babysit it afterward
14% of context to CLAUDE.md is the right trade. you're buying coherence across sessions, not burning budget. agents without it are just stateless functions with a chat interface.
Setup and workflow really change the results
the 14% lost to CLAUDE.md setup is the tax nobody talks about. most people skip it and wonder why their results suck. it's like trying to cook without mise en place
most folks skip the setup hoping for magic instead of work. another 25 step guide proving you're wrong is exactly what the algorithm wants
Wow you are CNBC and can’t afford a decent mic/ audio editing
90% of Anthropic's own code written by Claude and most users haven't touched a single config, that gap explains everything
90%代码自己写，有点夸张吧
Taking time to set context around my workflows and asking Claude to update it as I go has been a game changer for me.
Claude code is a buggy pile of trash though.
Context Engineering Is Replacing Prompt Engineering. Here's How It Works:
コメント：クロードのセットアップをスキップしている人が多い理由を説明しました。実際に結果を得られない人がほとんどいるのは、クロード本体によって書かれたものの90％のコード、CLAUDE.mdで書く前に失われる14％、弁護士、デザイナー、金融関係者が実際のものを作っていること、CS学位もバックグラウンドも何もないことです。経験者が数ヶ月間クロードを使用しても、1つも設定を変更していない場合、全てが変わる20ステップをスキップしています。おそらく25ステップ。
14% stat is the key — most devs treat Claude like a search bar instead of giving it system-level context and tooling. Setup quality directly correlates with output quality.
90% stops being disputed after you've run Claude on multi-step tasks unsupervised for a week.
claude's self-coding is next level
this is Awesome.
setup is everything no doubt
25 STEPS !!!!! How bought the frigen AI figures out it's not "set-up" correctly on it's own and asks the questions to do so. OpenClaw didn't have this problem ? Stop shilling an internal tool as a production end user product.
クロードコードの創設者であるボリス・チェルニは、ほとんどの人々がクロードから本当の結果を得ていない理由を説明しました。それによると、大部分の人々はセットアップを全くスキップしてしまっています。経営での90%のコードはクロード自身が書いており、CLAUDE.mdを書く前に14%が失われます。弁護士、デザイナー、および財務担当者がそれを使って実際のものを構築しているとのことです。CS学位やバックグラウンドは必要ありません。数ヶ月間クロードを使用しているのに1つも設定を変更していない場合、全てが変わる20ステップをスキップしています。その代わりに今夜の別の番組ではなく、この動画を見てください。25の機能が全て含まれているので、フィードで迷子になる前にブックマークしておいてください。
man is always dropping claude bangers
Intressant!
curious what 'the 14% you' actually finishes as. prompt scaffolding vs actual code review is a huge gap, what was the full claim?
How to do steps should be in a skill file.
me and you love to listening Boris Cherny
