Video summary
Are we cooked? Claude Fable is out of hand.
Main summary
Key takeaways
What the video set out to test
-
A “vibe coding” vs “senior developer” style prompt/AI workflow challenge: recreate a specific animated UI button (from a Twitter/X post by Raul Drunka) using AI/code generation.
-
Two builders compete in a timed format:
- Wes: 10-minute build + 10-minute check-ins, using AI/vibe coding and voice dictation
- Randy (“producer Randy”): newer to coding, mostly video/script automation, using AI-driven text-to-code workflows
- End scoring is based on:
- Aesthetics: how closely the replica matches the original button video (visuals + animations)
- Code quality/scalability: whether AI-generated code is maintainable and works across sizes/responsive scenarios
- Deployment/Interactivity: final deliverable must be a public URL where the button is usable
The UI/product features they implemented (key technical concepts)
They reproduced a Windows-like monochrome blue mosaic button with multiple layered effects.
Mosaic background generation
- A JavaScript canvas approach to generate a grid of square tiles (“mosaic”).
- Parameters explicitly referenced:
- tile size
- gap
- grid dimensions
- blue color ranges
- In Randy’s case, the implementation becomes very heavy DOM-based (called out during evaluation).
Hover behavior
- Hover causes the button to lighten/illuminate, affecting:
- the surface
- the grid lines between tiles
- Includes a white/glow component and a more complex feel described as:
- “yaw/skew”
- side-to-side skewing
- subtle 3D tilt that changes with cursor position
- They iteratively adjust:
- timing
- brightness
- translucency
- easing to match the original
Mouse-position-driven effects
- Tracks cursor position and injects X/Y values into CSS variables to drive effects.
- Uses absolute positioning for the ripple/glow layer rather than advanced layout features.
- Mentions potentially using CSS anchor position, but defaults to simpler methods.
Click animation
- On click, a ripple/waterlike wave radiates outward from the center, with tweaks to:
- opacity
- width
- speed
- Also adds a secondary “explosion”/ring-like glowing element, gated to the click event.
3D illusion
- Uses CSS 3D transforms including:
- perspective
- preserve-3D
rotateX/rotateY-style behavior to tilt based on cursor corner
- Iteration problems included:
- jitter
- “jerky” motion
- rendering artifacts (e.g., too many decimals / pixel rounding)
Layering approach
A recurring theme: the look comes from stacked layers, such as:
- base mosaic colors
- top overlay/glow that dissolves or blends away on hover
- inner shadow / embossed effects
- radial gradient/vignette-like shading
Tooling/workflow used in the build
- AI code generation in a local repo using a starter template (HTML/CSS modified via prompts)
- Voice dictation to prompt the AI (speech-to-text mentioned as a workflow booster)
- Claude Desktop / Cloud CLI concepts referenced (Claude specifically mentioned)
- Fable used as the dev server/live preview system
- Deployment/publishing:
- Cloudflare
- Claude artifact page used for quick “public links” for demos
- sharing links via artifact page permissions/access controls
Major evaluation/iteration problems observed
Prompting limitations / “iteration hell”
- Small UI details (e.g., exact hover behavior per cell, ripple decay, vignette distribution, shadow geometry) are hard to describe precisely in words.
Rendering/performance quirks
- Base64 image injection for the mosaic caused rendering issues. Switching to canvas blob URLs was proposed (and partially fixed).
- Hover/click transitions produced artifacts/jitter until:
- math was adjusted (e.g., rounding)
- transform setup was corrected
Hover “settling back”
- Wes’s version reportedly did not fully revert on mouse-out (hover effect lingered).
Border radius fidelity
- “Squirle” was mentioned as a style request (interpreted as nonstandard rounded corners).
- Wes allegedly didn’t achieve the intended “squirle” behavior.
- Randy also struggled to match it in Firefox, with browser differences impacting results.
Scoring outcomes (final judgments)
Wes’s score: 3.5 / 5
- Positives
- Button concept mostly there
- Some effects present
- Negatives
- hover doesn’t revert properly
- missing/insufficient top bevel / protrusion feel
- tilt too exaggerated and not subtle/eased enough
- hover looks more like a flat white box than “reveal/see-through” over the mosaic layers
Randy’s score: 4 / 5 (wins by 0.5)
- Positives
- closer resting state
- strong animation feel on hover overall
- better match and polish
- Negatives / tradeoffs
- background implementation becomes extremely DOM-heavy (quoted as ~660
<i>tags with inline styles) - browser/render differences around “squirle” rounded-corner behavior
- some color/brightness mismatches (top too white; hover mouse area too bright/wrong composition)
- background implementation becomes extremely DOM-heavy (quoted as ~660
- Overall verdict: more accurate and polished despite less “clean” code structure.
Cost comment (for fun/credibility)
- Randy humorously mentions $745 to get the button via agent usage (framed as compute/tool cost).
Main speakers / sources
- Wes Boss — builder; senior-style prompt approach using Claude + vibe coding + voice dictation
- Randy Recctor / Producer Randy — builder + video producer/editor; uses AI scripting/workflows
- Syntax producer / host voice — questioning/directing the challenge; gives grading prompts
- Source UI reference: Raul Drunka — the original Twitter/X button video they recreate