Practical AI for work, content & online business. Testing tools, workflows and ideas. Sharing what works - and what doesn't
Czech Republic
Joined October 2026
- Tweets557
- Following110
- Followers62
- Likes1.2K
AI Workflow Lab retweeted
I gave AI a brief for a jet engine that takes itself apart. It came back as 455 parts, all built in code.
Click a button and the fan, compressor, combustor, turbine or nozzle slides out. At the end the whole engine flies apart on a turntable and snaps back together in under a second.
The full method is in the article below. Change the object in the brief and you get a new build.
0 to 7.6 m/s in 3.3 milliseconds. That’s the tongue of a robotic chameleon Claude built for me in code
Claude followed the method from this article: write the model before drawing a single part. Here the model is one spring: 900 N/m, a 4 g tongue, a 1.6 cm stroke. The 367 g peak, the 26 cm reach in 37 ms and every gauge on screen come from those three numbers, so the panels and the animation can’t disagree
Then the teardown, one button per part. The eye turrets pop out and split into 3 lens elements, 30 colour plates lift off each flank, the tongue fires at 1/20 speed, 20 tail segments uncoil on one tendon and the gripper toes let go of the rail. At the end it all comes apart on a turntable, and the last strike goes straight at the camera
No 3D models and no images. It’s one HTML file
AI Workflow Lab retweeted
A Chinese plant building AI servers for big tech assembles every unit in Blender first, and keeps finding screws nobody can reach
A design gets approved as a finished object, so nobody checks if the parts can actually go in one after another.
Claude takes the CAD, rebuilds the server in Blender and runs the assembly the way the line would:
> Order: every part goes in the sequence the line will use
> Reach: each tool path is checked against what's already installed
• What the dry run keeps finding
> A screw sealed behind a bracket that went in first
> A cable that only routes if the fan comes out
> A heatsink that needs three hands to seat
• What it changed on the floor
> The build order gets fixed before unit one, not unit one hundred
> Stations stop halting to call engineering
> New designs arrive with the sequence already proven
A full server goes through its virtual assembly in about seven minutes.
The worst one passed every design review and still stopped a line for two days.
One bracket had to go in before the screw it ended up covering.
A design that looks finished isn't the same as one you can build.
The whole dry run, part by part, is in the article below ↓
AI Workflow Lab retweeted
I asked Claude for a robot gecko that takes itself apart. It came back as 1,683 parts, every plate and toe written in code.
what you’re watching:
- it walks in, powers down and falls apart onto the floor
- the parts line up in rows and rebuild it one category at a time
- it swings its tail and opens its own skull to show the neural core
- it blows apart into a cloud of plates and snaps back together
- it grips the floor with all 20 toe pads, then runs off screen
Tokay geckos really do have 5 sticky toe pads on every foot, so I gave each one its own glowing part.
which animal should come apart next?
AI Workflow Lab retweeted
Fable 5.2 made a Japanese-inspired fps game in three.js on top of my sakura valley scene.
Less than 10mb html with no image or model files. Pretty nuts.
Inspiration: xitter.gq/Grummz/status/21202550…
My sakura scene: threeui.com/three-js/sakura-…
Fable 5.2 made this in 2.9 hours.
Everyone is doing this game challenge by @felixrieseberg. So I tried it myself (see vid).
Everything is built by Fable, all models, textures and sound, lighting etc. ZERO stock assets (with one obvious easter egg exception).
Mine took 2.9 hours and 30% less tokens using my Kun-Zhi engine to reach a comparable stage.
Could have saved more, but I used our Khorvis Blender plug-in to have Fable build the assets there which bumped token usage but improved textures and models.
He one-shotted his, mine took 2 stages (I had my effort on high and it quit early, so I told it to "make it more AAA with volumetric light effects" and raised effort to Extra).
The prompt: "Build me a FPS in the Kun-Zhi engine. Call it "Kun-Zhi-Line". Insanely detailed quality pls, something that looks visually like it can keep up with AAA. Make it small in scope (like a tiny level) but otherwise impressive photorealistic graphics."
While you can one shot simple games for web, using an engine like Kun-Zhi will greatly speed production, reduce costs, and produce a more robust, production ready game.
Kun-Zhi will be open source and available to Members in alpha state late Jan. Sign up for more info at KunZhiEngine.com
My 3D article says this method works badly for animals. So I asked Claude for a penguin
10 parts, all basic shapes: stretched spheres for the body, belly, head and flippers, one cone for the beak, one ring for the scarf. Claude wrote the script, ran Blender on my PC and even made it hop, flap and blink at the end
Fur, feathers and real anatomy are another story. For those I’d still sculpt by hand
AI Workflow Lab retweeted
This animal eats by pushing its stomach OUT of its body🤯
No brain. No blood. Instead, it pumps seawater through its own body to move.
So I built one in code, on a dark glass stage. And you can poke it:
> 560 hydraulic tube feet that actually step and grip
> Touch an arm and it curls away
> Cutaway: see the water-powered “plumbing” inside, from the ring canal and radial canals to the digestive glands
> Feed it and it wraps a mussel in its own stomach
> Break off an arm and watch it grow back (time-lapse)
> Lights out: plankton glows wherever it moves
No 3D models. No textures. No animator. One HTML file.
Strangest animal I’ve built so far.
Your turn: the article in the comments shows exactly how to build your own creature. Drop what you made below 👇
AI Workflow Lab retweeted
You breathe about 22,000 times a day and never think about it. So I had Claude rebuild your lungs as a machine: 747 parts, all of them code.
The anatomy is real. The right lung has 3 lobes, the left has 2 and a notch where the heart sits, and 18 rings of cartilage hold the trachea open.
what you’re watching:
- the parts fly in from a flat pack and lock onto the stand
- the bronchial tree lights up from the trachea down
- it breathes at 15 a minute (a normal resting rate is 12 to 20), and the diaphragm drops on every inhale
- one breath in slow motion
- the plates open up so you can see the airways inside
At the ends of those branches sit about 480 million alveoli, which is where oxygen gets into your blood.
Which organ should I rebuild next?
AI Workflow Lab retweeted
holy f*ck, 2027 is coming for web designers and most of them have no clue
the people learning AI tools right now are pulling ahead. everyone else may end up fighting for work they never had to fight for.
7 GitHub repos every designer should save before it's too late:
1. Open Design
107k stars. a full AI design studio that runs on your own machine. give it a brief and your coding agent builds a prototype that follows your design system, then your team leaves feedback right on it
github.com/nexu-io/open-desi…
2. Impeccable
108k stars. gives Claude a design vocabulary. type /impeccable bolder the pricing cards and it knows exactly what to do. 62 rules catch AI slop before it ships
github.com/pbakaus/impeccabl…
3. HyperFrames
83k stars, made by HeyGen. your agent writes HTML, it renders a real MP4. animated intros and promos from one prompt, no editing app
github.com/heygen-com/hyperf…
4. designlang
point it at any website and it pulls the whole design system: colors, fonts, spacing, motion, hover states. one command, 17+ files, plus a "slop score" out of 100 for your own site
github.com/Manavarya09/desig…
5. Anatomy
26k stars. ask Claude to explain anything and it builds a moving isometric machine that shows how it works. dozens of shaded parts, it reacts to your mouse
github.com/wheresryan22/anat…
6. Figma Context MCP
16k stars. paste a Figma link, your AI reads the real layout data instead of guessing from a screenshot. way closer to the original on the first try
github.com/GLips/Figma-Conte…
7. video-shotcraft
18k stars. cinematic product promos straight from Claude Code. 124 shot recipes, 150 motion previews, cuts synced to the beat
github.com/Vincentwei1021/vi…
the cell in this video?
one person, Claude Code and a few tools from this list. no agency, no team, no After Effects.
want to build something like it?
like + save this post, comment "CELL" and I'll send you the exact prompt.
Fable 5.2 built me a cyberpunk game from one seed: 2099
You fly a courier car through a night city of 275 towers. Grab a data core at the lime beacon, get it to the uplink tower, and stay away from the corp drones. Their red beams fill a trace meter, and a full trace burns a third of your shield. There is an onboard AI called VEX that warns you when to boost
Zero assets. Every tower, window, holo ad, drone and raindrop is generated in code, and the synthwave track is synthesized from a list of notes. No engine either, it’s one HTML file.
I asked Fable 5.2 for a video with music. It built the instrument first
It’s a marble machine in 3D. A sloped belt carries 244 marbles toward the edge, and each one drops onto a metal bar, a bass bar, a drum or a cymbal. The belt works as the sheet music, so you can read the next 3 seconds of the song on it before you hear them
There is no audio file anywhere. The sound is synthesized from the same list of 244 notes that moves the marbles, which is why every note lands on the exact frame of its impact. 13 bars at 100 BPM, 4 chords, and a run of 60 marbles in 9.6 seconds near the end
No samples, no 3D models, no libraries. From my message to the finished video took 45 minutes, and about 30 of those were rendering
AI Workflow Lab retweeted
Two stick figures. One frozen lake. Only one of them walks away.
I asked AI for a stickman fight. What came back has actual fight choreography
→ a head-on clash that turns into a shoving match of sparks
→ a four-hit exchange, every strike blocked
→ a low sweep, answered with a flip right over the opponent’s head, in slow motion
→ a kick that sends him sliding across the ice until it cracks
→ one final crossing strike, a white flash… and a blade that flickers out
All of it is code. Every pose, every spark, the reflections in the ice, the aurora, the snow, even the sound: the hum of the blades, the clashes, the ice cracking.
No animator. No keyframes drawn by hand. No stock footage.
Built with Claude.
Who wins the rematch: white or red?
AI Workflow Lab retweeted
THIS IS ACTUAL F*CKING GOLD
Anthropic just shared one of the best tricks for Claude Code
The idea: run Sonnet, Haiku and Opus 5.5 together, each one doing only the job it's best at
claude --model sonnet --advisor opus
Sonnet 5.5 runs the session and writes the code. Opus 5.5 stays on call as an advisor and reads the whole transcript every time it's consulted.
opus-level judgment on the big calls, without paying opus rates for every routine turn.
claude decides when to ask. per the docs it usually happens at 3 points:
> before it commits to a plan
> when the same error keeps coming back
> before it calls the task done
add Haiku 5.5 subagents for the grunt work and you get the full tree:
> Sonnet 5.5: main session, edits code
> search on Haiku 5.5: maps the repo
> grep on Haiku 5.5: finds call sites
> docs on Haiku 5.5: pulls the API reference
> Opus 5.5: advisor, only at the decision points
anthropic's own numbers from the advisor launch: sonnet with an opus advisor scored 2.7 points higher on SWE-bench Multilingual than sonnet alone, and cost 11.9% less per task
paste this into Claude Code:
"set up my Claude Code like this:
1. main session on Sonnet 5.5, Opus 5.5 as advisor (advisorModel: opus in my user settings)
2. three read-only Haiku 5.5 subagents in .claude/agents/ (model: haiku): search maps the repo, grep finds call sites, docs pulls the API reference. tools: Read, Grep, Glob only, docs also gets WebFetch
3. one rule in CLAUDE.md: consult the advisor before locking a plan, after the same error twice, and before calling a task done
show me every change as a diff first. no edits until i say go."
I asked Claude to take apart a drone that doesn’t exist, with one rule: every number on screen has to come from one formul
The Kestrel K-4 is made up: 249 g, 4 motors, a 2S battery. Claude wrote the physics before it drew a single part. Throttle sets motor speed, speed sets thrust and power, power drains the battery. Every gauge reads that one model, so at 14,000 rpm the thrust lands on 2.44 N, the drone’s own weight, and it lifts off
Then it films the teardown in 26 seconds. The propellers come off, the top shell lifts over the flight controller and GPS, the battery slides out, the gimbal camera sweeps, and the motor bells pop off to show 12 copper windings and 14 magnets
No 3D models and no keyframes. It’s one HTML file, built with the method from this article
AI Workflow Lab retweeted
give FABLE 5.2 a 3-view drawing and it sculpts your game boss in blender by code FOR FREE
380 steps and no 3d generator
> the only input was the concept sheet: front, side and back
> it blocked him out of boulders, grew root arms, a face and vines, then refit the silhouette 20 times
> front, side and back now match the drawing at 0.89 to 0.90 iou
> stone cracks, bark and moss are shader nodes, and the 7,701 plants are modelled and scattered by code
> 1,675 lines of python, about 32 minutes of fable time
try it with your own 3-view sheet, the full character pipeline from drawing to rigged model is in the article below
