茶谷 和弘 / K.Chatani

AIプラットフォーム開発•コンサル スタートアップ「Coo Quack」 代表技術統括 / CTO、趣味は、ランニング、グルメ、映画鑑賞。新しい事に常に挑戦します！

[ZEUS⚡️ @zeuuss_01]

FABLE 5 + HIGGSFIELD CAN BUILD A $35,000 ANIMATED WEBSITE. IN ONE AGENTIC SESSION. FOR ~$12 IN CREDITS.

stop paying a web studio $6,000-$35,000.

stop wiring GSAP, Lenis, and frame extraction by hand.

Claude Code writes it. Higgsfield renders it.

WHAT THE BUILD PRODUCES:

→ a fully animated, scroll-driven site
→ cinematic motion clips from 30+ generative models
→ GSAP ScrollTrigger timelines - zero hand-coded

keyframes

→ Lenis smooth-scroll, tuned pacing
→ automated frame extraction + asset optimization
→ six cinematic effects baked in: film grain, particles, vignette, glass cards, color tints, scroll pacing
→ responsive layout + copy

THE STACK:

→ Claude Code - concept, scaffolding, scroll code, QA
→ Higgsfield (MCP) - hero clips, transitions, ambient loops, thumbnails
→ GSAP + Lenis - the motion layer, written for you

CONNECT HIGGSFIELD (MCP):

add it as a custom connector in Claude Code:
mcp_servers:

higgsfield:
url: "https://t.co/ccj3bJ9926"

one OAuth flow. done.

now Claude can generate and pull clips directly - no manual exporting.

WHAT TO PROMPT:

concept + scroll:

"read this brief, script the scroll - what the visitor feels at second 3, 15, 40. scaffold the site with GSAP ScrollTrigger + Lenis."

motion assets:

"generate the hero sting and one b-roll clip per section from the story. 3–5s, high-res."

polish pass:

"bake in film grain, particles, vignette, glass cards, color tints, scroll pacing. no config."

QA:

"check load speed, mobile breakpoints, and whether the scroll actually lands. rewrite whatever doesn't."

WHAT THIS REPLACES:

→ web studio build: $6,000-$35,000+
→ motion artist: $800-2,000/project
→ front-end dev: $2,000-10,000/project
→ weeks of handoffs: gone

Fable 5 + Higgsfield: a subscription + a few dollars of credits. one session.

SETUP IN 10 MINUTES:

- install Claude Code
- add the Higgsfield MCP + authenticate
- drop your brief + references
- let it scaffold, generate, and animate in one pass
- preview, send fixes in plain English, ship

the pipeline was the moat. it just became a prompt.

Follow me, comment "WEB" and I'll send you the full step-by-step Playbook.

full breakdown in the article 👇

[茶谷 和弘 / K.Chatani @chataclaw]

NetflixエンジニアがClaude APIコストを95%削減するツールを公開。ログとJSONを圧縮しつつロジック精度100%維持、データ完全ローカル保存。AIエージェントのコストパフォーマンスゲームチェンジャー

[Viktor Oddy @viktoroddy]

Premium 3D Animated website

❤️‍🔥 Fable 5 + Claude Code → Build

Prompt ↓

[Anatoli Kopadze @AnatoliKopadze]

Godfather of AI: "If you sleep well tonight, you may not have understood this lecture."

This 47-minute lecture is the best thing I've seen about AI in the last few months.

Hinton built the neural networks behind every AI alive, then quit Google to warn us it's already ahead of us on most cognitive tasks.

Despite that, most people open Claude, type one thing, close the tab and think they're using AI, but they're using maybe 10%.

I turned his talk into 17 Claude features 99% of users never find.

Watch the lecture, then read the article below.

[茶谷 和弘 / K.Chatani @chataclaw]

Claude Fable 5が45分でBアプリをクローン。コードを書く前に10分だけ「考える」だけで完全動作クローンが完成。この開発効率化はデバックログを完全無用化する。開発パラダイムの根本転換

[茶谷 和弘 / K.Chatani @chataclaw]

NetflixエンジニアがClaude APIコストを95%削減するツールを公開。ログとJSONを圧縮しつつロジック精度100%維持、データ完全ローカル保存。AIエージェントのコストパフォーマンスゲームチェンジャー

[Ruben Hassid @rubenhassid]

How to prompt (the new) Claude Fable 5:

1. Task

☑ Give the goal + the why, not the steps:
"I'm working on [LARGER GOAL] for [WHO IT'S FOR]. They need [WHAT THE OUTPUT ENABLES]. With that in mind: [TASK]."

No "write me a follow-up email."
↳ That's a task. Claude 5 works well on goals.

2. Context

☑ The Project knows, the chat forgets:
"Use everything in this Project first: my about-me doc, instructions, uploaded examples. If it's not there, pull it from my connectors - don't guess."

Stop re-explaining who you are every single chat. Put it in a Project once.

3. Skill

☑ Saved once, loaded in one word:
"/[skill-name] — apply it fully."
Anything you've explained to Claude twice should become a skill. That's the rule.

4. Effort

☑ Don't let it undersell itself:
"This is a [routine / hard / hardest-unsolved] problem. Scope it like it's at the top of your range."

Claude 5 is built for your hardest problems.
Testing it on easy ones undersells it.

5. Act

☑ Stop it from overplanning:
"When you have enough information to act, act. Weighing a choice? Give a recommendation."

No more 4 paragraphs of options it won't pursue.

6. Scope

☑ The simplest thing that works:
"No extra features, refactors, or abstractions. If I'm describing a problem, the deliverable is your assessment."

Smart models love to over-deliver. Cut it off.

7. Delegate

☑ Subagents do the boring work:
"Split independent sub-tasks across subagents & keep working. Verify with a fresh-context subagent."

Claude 5 manages its own team now. Let it.

8. Evidence

☑ Audit before reporting:
"Before reporting, audit every claim against a real result from this session. Unverified? Say so."

This one line nearly eliminates made-up progress reports. Anthropic tested it themselves.

9. Memory

☑ Promote lessons into the Project:
"When you learn something about me that will matter next time, tell me at the end."

You paste it into the Project instructions.
Claude gets smarter every week.

10. Checkpoint

☑ Pause only when it must:
"Pause only for: destructive actions, real scope changes, or input only I can provide. Never end your turn on a promise."

No more "Shall I proceed?" every 3 minutes.

11. Report

☑ The TLDR comes first:
"Open with the outcome. Complete sentences, no shorthand I never saw. Clear beats short."

You get the answer, not the long journal.

Copy the full template here:

Step 1. Subscribe for free → https://t.co/psB7XxB2Y4.
Step 2. You will have two choices: free or paid.
Step 3. Choose the free tier. Don't pay for anything.
Step 4. Open your welcoming email (wait 30 sec).
Step 5. Access my entire prompt library & skills.
♻️ Repost this to help others.

[AI Guides @free_ai_guides]

Your agent keeps failing, and you keep rewriting the prompt. The fix is 3 layers above where you're looking.

This map breaks down the 4 layers of AI engineering: Prompt, Context, Harness, and Loop. Each one wraps the previous, and most builders never get past layer 1.

Save the map, then read the guide below.

[Miles Deutscher @milesdeutscher]

I don't prompt Claude Code anymore.

I have loops running that prompt Fable, and my job is just to write loops.

This is the Boris Cherny method, and I have to say, it's extremely powerful.

Everything you need to get started with loop engineering (as a complete beginner):

[Alex Prompter @alex_prompter]

There are layers of working with AI. Prompt engineering. Context engineering. Harness engineering. Loop engineering.

Layer 1 is prompt engineering. What you type into the chat window. How you word the instruction, what you ask for, what you tell it to avoid. This is where everyone starts. It matters, but few people go further.

Layer 2 is context engineering. Everything the model sees before your prompt. System instructions, reference files, conversation history, examples of good output. A mediocre prompt with great context beats a great prompt with no context every time.

Layer 3 is harness engineering. The code around the model. Tool routing, verification steps, retry logic, structured outputs. This is what makes AI reliable instead of just impressive. When the model checks its own work before returning it to you, that's the harness.

Layer 4 is loop engineering. The system runs itself. You set a goal and a stop condition. The loop prompts the model, checks the output, adjusts, and repeats without you. This is where you stop being the bottleneck entirely.

Each layer wraps the one before it. Better prompts help, but without context the model guesses. Context helps, but without a harness the output is inconsistent. A harness helps, but without a loop you're still manually triggering every run.

Full setup guide for all 4 layers using Fable 5 below.

[茶谷 和弘 / K.Chatani @chataclaw]

Sonnet 5が日常用途でデファクトになりそう。Opus 4.8との使い分け戦略が鍵だな。https://t.co/tUiy99ptPw

[こすもす｜AIで起業しろ @cosmos_hzokujin]

GitHubで話題の「claude-video（約3K Star）」。

これ、単なる要約ツールではなく「動画の作り方」を根本から変えるゲームチェンジャーです。

URLを渡すと、AIが動画をコマ単位で分解し、音声も文字起こしします。
推測ではなく、物理的に動画を「見て・聞いて」いる状態になります。

ただ、動画を要約させて満足していては意味がありません。 真の価値は、バズ動画の「冒頭3秒のフック」「カメラワーク」「リズム」をAIに分解させ、【構造化されたアセット（資産）】に変換すること。

そして、その抽出したバズの骨組みを再びAIに投げ、「この構成で、うちの商品用の台本に書き換えて」と指示を出す。

ここまでやって初めて、属人的な「勘」や「センス」に頼っていた動画制作が、再現性のある【エンジニアリング（構築作業）】に変わります。

もちろん最後は人間の直感も必要ですが、「なんとなくバズりそう」という当てずっぽうな時代は終わりました。

優れた構成をAIで抽出し、転用する。 コンテンツ運営は間違いなく、エンジニアリングの領域に入ってきています。
https://t.co/ARj26iE6dt

[Nico @nicos_ai]

El Loop Engineering es el siguiente paso después del Prompt Engineering

La mayoría de las personas aún usan Claude Code, Codex o Cursor como una chatbot:

1. Prompt
2. Esperar
3. Copiar
4. Corregir
5. Prompt otra vez

Este repo muestra el siguiente paso:

Dejas de hacer prompts al agente.

Diseñas el loop que le hace los prompts al agente por ti.

Incluye:

→ Loops de triaje diario
→ Loops de niñera de PR
→ Loops de barrido de CI
→ Loops de barrido de dependencias
→ Loops de redacción de changelog
→ Loops de limpieza post-merge
→ Loops de triaje de issues

También te da CLIs para:

• Escalar un loop
• Estimar el costo de tokens
• Auditar si tu repo está listo
• Agregar memoria/estado
• Agregar handoff humano
• Agregar gates de verificación
• Ejecutar agentes de manera segura a través de GitHub Actions

El Prompt Engineering se trataba de escribir mejores instrucciones.

El Loop Engineering se trata de construir un sistema donde los agentes siguen:

trabajando, verificando, corrigiendo y escalando sin que tú estés cuidando cada paso.

Esto es lo que parece la codificación con IA cuando deja de ser una sesión de chat y empieza a convertirse en un sistema operativo para equipos de software.

Enlace abajo👇

[茶谷 和弘 / K.Chatani @chataclaw]

遅すぎると言われ続けたE-ink画面が、Claude Codeの出力ペースと組み合わさった瞬間に化けた。開発者はreMarkable PaperにSSH接続してコーディング環境を作り上げた。これは「ペーパーレス」ではなく「ペーパースペース」の時代の到来だ

[Oikon @oikon48]

Claude Code 2.1.200 (抜粋)

- `AskUserQuestion` ダイアログの挙動を変更し、デフォルトでは自動継続しないようにした。`/config` からアイドルタイムアウトをオプトインで有効化できる

- CLI・`--help`・VS Code・JetBrains 全体で「default」permission mode の表記を「Manual」に変更。`--permission-mode manual` および `"defaultMode": "manual"` も `default` と並んで受理される

- テキスト出力を一切生成する前にレート制限で打ち切られたサブエージェントが、失敗として扱われず空の結果を返してしまう問題を修正

- バックグラウンドエージェントの出力に含まれる制御バイトがエージェントビューの端末にそのまま到達してしまう問題を修正

- 同一リポジトリの git worktree からプロジェクトスコープのプラグインが正しくロードされない問題を修正

- 音声を録音できなかった際に、音声ディクテーションが誤解を招く「Voice connection failed」メッセージを表示する問題を修正

- インストールスクリプトを改善し、システムのメモリ不足によってインストールが強制終了された場合にその旨を説明するように

https://t.co/3bnEjPxcNC

[茶谷 和弘 / K.Chatani @chataclaw]

AnthropicのAI創薬プログラム開始、科学者向けツールに加えて社内で本格的な創薬に挑戦。医療AIの次のフロンティアが動き出した https://t.co/0mvuE7vJtr

[Deliriumm @0xDeliriumm]

ANTHROPIC DROPPED THEIR OWN OBSIDIAN BRAIN THAT MANAGES ALL COMPANY INNOVATIONS AND KNOWLEDGE FLOWS

00:22 8,893 nodes, 4,729 links, 8,893 views - a knowledge graph so dense it looks like a galaxy when you zoom out

Marginalia Collection, Glossary Backbone, Comparative Grammar MOC, Oral History Transcripts, Chinese-English Translation Magazine founded 1973 - every cluster its own universe of connected knowledge

Master Index in the center with 9,000+ documents and dozens of cross-references pulling everything together into one navigable system

Field Notes Archive, Translation Studies Index with 419 cross-references, Survey Data pool with 212 coded entries - this is not a second brain, this is an entire research civilization

Fable 5 came back globally on July 1 2026 - and before you run the same prompts you used on Opus 4.8, Anthropic published an official prompting guide because old prompts don't hit the same

the company building the most powerful AI in the world uses an Obsidian graph to manage its own knowledge - and now showed you exactly how to build yours 🧠

[Vaibhav Sisinty @VaibhavSisinty]

Someone just open sourced the most complete Fable 5 agent orchestration workflow I have come across.

And the main lesson flips how most people think about AI coding agents. 🧵

https://t.co/ONbjYPM65x

[茶谷 和弘 / K.Chatani @chataclaw]

最近のAIエージェント開発トレンドは「プロンプティング」から「ループ設計」へ。Fableを使った自動研究の実例がすごい