I spent $1,000 on Claude Fable 5 (Mythos), it's incredible
A 14-minute first-impressions report on the best coding model available — and the 12-day window before it stops being free.
June 11thA 22-minute live-demo tutorial showing how design tokens and Claude Design eliminate the inconsistency that makes AI-built apps look cheap.
The reason AI-built apps look inconsistent is not bad prompting — it is the absence of a design system with tokens, and Claude Design solves this in one screenshot-to-system pipeline.
AI-built UIs get inconsistent because developers skip the design system step. The fix is a four-stage pipeline: gather a reference site you like, feed it to Claude Design to auto-generate a token-based design system, spend time tweaking colors and components until the system feels right, then use Claude Design's built-in Send to Claude Code feature to implement it in one shot. Once tokens are baked into the codebase, every future AI coding session inherits them automatically — so fixing a spacing issue fixes it everywhere, and adding dark mode no longer breaks half your components.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Create a free account →
Hook that names the AI smell in UI, then demonstrates the inconsistency in the author's own Pluto app — different button radii, dashboard vs landing page mismatch.

Defines what a design system is and why tokens (semantic variable names instead of hardcoded values) are the core mechanism for consistency, theming, and centralized maintenance.

Twitter bookmarks via GatherOS, Mobin, and reference sites. Uses family.co as the demo inspiration source.

Screenshots family.co, uploads to Claude Design, prompts it to generate a token-based design system with components on one page.

Scriba learning platform for AI engineering — agents, MCP, scalable multi-tool systems.

Reviews the generated Family Design System and uses the markup tool to remove confetti decorations and the AI logo. Also shows the Vercel design.md approach as an alternative starting point.

Asks Claude Design to generate dialog, form, checkbox, dropdown, tooltip, and step components. Reviews the expanded system before handing off to code.

Spins up a SvelteKit app. Uses Share to Send to Claude Code to export the design system. Claude Code imports via MCP and builds landing page, auth page, and dashboard in one prompt.

Reviews the three generated pages with consistent tokens throughout. Introduces BUI for micro-animations layered on top of the design system.

Recaps the four-step workflow and emphasizes adding your own aesthetic input — use AI as a tool, not a replacement for taste.
When tokens are established before coding begins, every AI session that touches the app inherits the same visual rules — fixing one component fixes all of them.
“You can look at a web or mobile app, and you can tell if it's just fully been designed with AI because it has that smell.”
“If you use AI to design the entire thing, no sauce or input from you, your site's gonna look like everybody else's.”
“Because you have a design system, once you fix one thing, it affects the entire thing. The fix is applied everywhere.”
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
You can always tell when a web app was designed entirely by AI. It has that smell — weird borders, a landing page that looks polished and a dashboard that looks like a different product entirely. In this video, one developer shows you why the fix is not better prompts, it is the step you have been skipping: a token-based design system built before you write a single line of app code.
Sequential pipeline that converts visual inspiration into a token-based design system and then into consistent app UI.
The core argument for why tokens are worth the setup cost, especially when building with AI.
“We're about to hit 100k subs. You should be subscribing if you're watching right now.”
Mid-video aside during a demo pause, delivered casually while the AI is still running. Organic, not a hard pitch.
00:00
00:23
00:37
00:54
01:19
01:31
01:47
02:04
02:21
02:37
02:55
03:09
03:33
03:44
04:02
04:16
04:26
04:43
05:01
05:23
05:47
06:00
06:13
06:38
06:53
07:03
07:20
07:33
07:49
08:07
08:27
08:43
08:59
09:16
09:33
09:47
10:06
10:22
10:39
10:56
11:12
11:36
11:46
11:57
12:19
12:35
12:55
13:09
13:25
13:43
13:58
14:15
14:31
14:51
15:07
15:24
15:38
15:56
16:07
16:33
16:45
17:01
17:18
17:34
17:51
18:02
18:31
18:41
18:57
19:07
19:24
19:44
20:06
20:15
20:33
20:54
21:10
21:20
21:44
21:58A 14-minute first-impressions report on the best coding model available — and the 12-day window before it stops being free.
June 11thA 13-minute breakdown of one builder's agentic engineering stack: three Claude Code skills, an agents.md file, and the token-math that explains why they are not the same thing.
June 9thRas Mic's argument for why a long conversation before plan mode beats plan mode alone -- and a live demo building a mobile companion app for his AI agent platform.
June 5thA 9-minute feature breakdown of the update that finally makes Claude Design usable in a real design-to-code workflow.
June 19thA 14-minute proof that the AI slop problem is a documentation problem, and that Fable only beats Opus when you let it make decisions from scratch.
June 11thAn 87-minute practitioner guide to which AI tool belongs at which stage of the design process, and why using the wrong one burns tokens and time.
May 4th