Загрузка видео...

Не удалось загрузить видео

На главную

I leaned into this opus 5.5 ultracode dynamic workflow thing with 9 subagents, just to see what it can cook up, this is the result Sink Village Three.js

48,442 просмотров • 18 дней назад •via X (Twitter)

Комментарии: 21

Фото профиля Vibefold
Vibefold18 дней назад

@threejs 9 subagents on one scene is the part i want to see broken down. who owned what: terrain, water, lighting? the split is usually where these either sing or fight each other

Фото профиля Louis Vuitton ×̷̷͜×̷ ₉⁹₉ 🌙🐈
Louis Vuitton ×̷̷͜×̷ ₉⁹₉ 🌙🐈18 дней назад

@threejs Wtf thats sick

Фото профиля Scott
Scott18 дней назад

@threejs 🙏🏼

Фото профиля Sloyd
Sloyd18 дней назад

@threejs This is simply amazing!

Фото профиля Scott
Scott18 дней назад

@threejs Thanks!

Фото профиля sw1
sw118 дней назад

@threejs The final boss is someone doing the dishes 😂

Фото профиля Victor
Victor18 дней назад

@threejs Kitchen topology is messed up tho. Nobody would place a sink there.

Фото профиля Scott
Scott18 дней назад

@threejs Exactly

Фото профиля urⓑanⓑor.ⓑtc | Bitmap |🟧
urⓑanⓑor.ⓑtc | Bitmap |🟧18 дней назад

@threejs love ittt

Фото профиля Ishu Agrawal
Ishu Agrawal18 дней назад

@threejs sheesh... how much of your usage did this take?

Фото профиля Scott
Scott18 дней назад

@threejs Less than 10% weekly, I’m on 5x max plan My usage resets tonight so I’m just burning it off

Фото профиля Ishu Agrawal
Ishu Agrawal18 дней назад

@threejs damn I can never get my outputs to look like this even with references

Фото профиля Tone
Tone18 дней назад

@threejs This isn’t a one shot right? What kind of back and forth did you have and didn’t describe what tools it’s using? I assume three.js but anything else notable? Dope result

Фото профиля Y
Y18 дней назад

@threejs the waterfall into the tub is great. how did the 9 subagents split it, by area or by layer (terrain, water, houses)?

Фото профиля awasemiso+
awasemiso+17 дней назад

@threejs うはっ、めっちゃ好き。 この発想は最高だ

Фото профиля Brick & Mortar
Brick & Mortar18 дней назад

@threejs artsy

Фото профиля Lucas
Lucas18 дней назад

@threejs thats so cool!

Фото профиля naresh
naresh18 дней назад

@threejs daymm , what's more interesting is the water animation and the desgin of it .. it looks so flawless

Фото профиля KLY正经人家
KLY正经人家17 дней назад

@threejs 这个洗手池太丰富了吧

Фото профиля 🇺🇸 To The Stars 🇺🇸
🇺🇸 To The Stars 🇺🇸17 дней назад

@threejs Ok, this is creative. I love the sink spout, lol

Фото профиля Doodle Dash
Doodle Dash18 дней назад

@threejs 9 subagents? Bro assembled the Avengers to build a village.

Похожие видео

THIS IS A PURE F*CKING GOLDMINE Anthropic just shared one of the best tricks for Claude Code. The idea: let Opus 5.5 write the plan as a script, and let a swarm of cheaper agents run it. Bigger jobs and way fewer Opus tokens, because the script holds the plan and Opus only sees the final answer. You just launch Claude Code like this: claude --model opus --effort ultracode here's the split: → Opus 5.5 turns your task into a workflow script → Haiku 5.5 agents scan every file in parallel → Sonnet 5.5 agents patch each fix in its own isolated copy → you stop burning Opus context on a hundred agent transcripts Opus only shows up at three moments: → writing the workflow before the run → reading the verified results after it → signing off before anything ships Opus plans. Haiku scans. Sonnet ships. JEV engineering is the same move one layer down: the forks that need no thinker (which file, which tool, retry or stop) go to JEV in milliseconds, and the big models only see the ones that split the full tree > Opus 5.5 runs the main session and writes the workflow > scan audits every file on Haiku 5.5 > edit patches each fix on Sonnet 5.5 in an isolated copy > verify cross-checks every finding on Sonnet 5.5 > Opus 5.5 reads only what survived verification > JEV routes micro-forks in milliseconds paste the tree and this prompt into Claude Code ↓ "Rebuild my Claude Code setup around this tree: 1. Run the main session on Opus 5.5 with ultracode on. 2. In every workflow, run read-only scan stages on Haiku 5.5 and edit stages on Sonnet 5.5, each edit in its own isolated copy. 3. Add one rule to CLAUDE.md: every finding gets an independent verifier agent before it's reported, and fix loops stop after two rounds with no progress. 4. Route fast deterministic forks (tool choice, path selection, retries) through JEV. Show me every change as a diff first. No edits until I say go." ↳ save it, it'll come in handy.

xsmryze

60,847 просмотров • 24 дней назад

THIS IS A ACTUAL F*CKING DIAMOND Anthropic just documented dynamic workflow orchestration for Claude Code The idea: let Opus 5.5 write the workflow script, swarm Haiku 5.5 scans at 340 tok/s, run Sonnet 5.5 isolated edits, and keep Fable 5.1 on call as the dedicated adversary: Dozens to hundreds of subagents with zero Opus burn: the workflow script holds the loop and Opus only sees the verified final diff You just launch Claude Code like this: claude --model opus --effort ultracode here's how it splits: → Opus 5.5 writes the workflow script to orchestrate control flow and state → Haiku 5.5 swarms repository files at 340 tok/s to find candidate targets → Sonnet 5.5 applies patches in isolated worktrees and runs test suites → Fable 5.1 validates the entire workflow as dedicated adversary on call → you stop burning Opus context on hundreds of intermediate subagent transcripts Fable 5.1 only steps in at three moments: → before the script runs: verifies subflow.js for runaway loops and missing bounds → when a patch fails twice in parallel: audits root invariants instead of patching symptoms → before calling it done: cold adversarial review of aggregated diffs before Opus sign-off Opus architects · Haiku scans · Sonnet patches · Fable audits JEV engineering is the same move one layer down: the forks that need no thinker (which phase, which tool, retry or stop) go to JEV in milliseconds, and the big models only see the ones that split the full tree > Opus 5.5 runs the main session and writes the workflow script > pipeline() fans out Haiku 5.5 to scan every file at 340 tok/s > parallel() spawns Sonnet 5.5 in isolated worktrees to apply patches > Fable 5.1 stays on call as adversary to audit scripts, failures, and final diffs > control flow stays in JavaScript so Opus context never bloats > JEV routes micro-forks in milliseconds paste the tree and this prompt into Claude Code ↓ "Rebuild my Claude Code setup around this tree: 1. Run the main session on Opus 5.5 with ultracode enabled via /effort ultracode 2. In every workflow, run read-only scan phases on Haiku 5.5 at 340 tok/s and edit phases on Sonnet 5.5 in isolated worktrees 3. Keep Fable 5.1 on call as dedicated adversary to audit workflow scripts, double-failure loops, and PR diffs 4. Add one rule to CLAUDE.md: every finding must be independently verified by Fable before reporting, and patch loops abort after two failed iterations 5. Route fast deterministic forks through JEV, show all changes as diffs first. No edits until I say go" ↳ save it, you'll use this every single day

mirku

21,682 просмотров • 15 дней назад

THIS IS A PURE F*CKING GOLDMINE Anthropic just shared one of the best tricks for Claude Code. The idea: let Opus 5.5 write the plan as a script, and let a swarm of cheaper agents run it. Bigger jobs and way fewer Opus tokens, because the script holds the plan and Opus only sees the final answer. You just launch Claude Code like this: claude --model opus --effort ultracode here's the split: → Opus 5.5 turns your task into a workflow script → Haiku 5.5 agents scan every file in parallel → Sonnet 5.5 agents patch each fix in its own isolated copy → you stop burning Opus context on a hundred agent transcripts Opus only shows up at three moments: → writing the workflow before the run → reading the verified results after it → signing off before anything ships Opus plans. Haiku scans. Sonnet ships. JEV engineering is the same move one layer down: the forks that need no thinker (which file, which tool, retry or stop) go to JEV in milliseconds, and the big models only see the ones that split the full tree > Opus 5.5 runs the main session and writes the workflow > scan audits every file on Haiku 5.5 > edit patches each fix on Sonnet 5.5 in an isolated copy > verify cross-checks every finding on Sonnet 5.5 > Opus 5.5 reads only what survived verification > JEV routes micro-forks in milliseconds paste the tree and this prompt into Claude Code ↓ "Rebuild my Claude Code setup around this tree: 1. Run the main session on Opus 5.5 with ultracode on. 2. In every workflow, run read-only scan stages on Haiku 5.5 and edit stages on Sonnet 5.5, each edit in its own isolated copy. 3. Add one rule to CLAUDE.md: every finding gets an independent verifier agent before it's reported, and fix loops stop after two rounds with no progress. 4. Route fast deterministic forks (tool choice, path selection, retries) through JEV. Show me every change as a diff first. No edits until I say go." ↳ save it, it'll come in handy.

delost

402,419 просмотров • 1 месяц назад