I Built 6 Websites in 17 Minutes With Claude Fable 5
A six-level stress test of Claude Fable 5 as a web design engine — from a two-minute landing page to a scroll-driven 3D experience that used to take hours.
June 11thOne prompt, a pre-written architecture repo, and a Claude-orchestrates-Codex workflow turn a static redesigned site into a fully autonomous CMS that writes, illustrates, and publishes its own articles.
A single context-rich prompt let an orchestrator AI model delegate the heavy coding to a second execution model and build a fully working, autonomously-publishing CMS backend without the operator writing or touching any code.
A designer who'd already migrated his site off Framer had no way to publish content — no CMS, no backend. He fed Claude (used here as an orchestrator he calls Fable 5) a pre-written backend architecture repo and told it to delegate execution to Codex GPT-5.5, giving it business context, not just instructions. The result: a Supabase-backed CMS with a Kanban content pipeline, scheduled trend scans, and an autonomous loop that writes, illustrates, and publishes SEO-optimized articles in his brand voice, gated by a manual-approval 'safe mode.' The lesson: pairing a strategist model with an execution model, and handing it a concrete architecture spec instead of open-ended instructions, is what turned a vague goal into a working system in one session.
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 →
Cold open: the front end was already migrated off Framer in a prior video, but with no CMS there's no way to publish. Today's goal is to get Claude to build the backend, then later a CRM.

Introduces 'digital sovereignty' — owning your own front end, back end, data, and analytics instead of renting platforms that dictate your workflow.

Switches to Fable 5, shown burning through weekly usage credits. Explains the workflow: Fable acts as strategist/orchestrator, delegating heavy execution to Codex GPT-5.5. Notes Anthropic's own guidance to explain 'why' when prompting, not just 'what.'

Pastes a prompt describing the desired autonomous publishing workflow, then pulls up the open-source 'Digital Home Backend Starter' GitHub repo and hands it to Claude with instructions to review and install it into the existing front end.

Argues pretty front ends are commodified now that AI can replicate them in one prompt; the resellable value is in backend systems (content, leads, data) — pitches his Skool community that teaches the full setup process.

Returns to find the autonomous publishing loop verified end to end — a full SEO/GEO-optimized article with a generated hero image and FAQ block already live on the site, written in his brand voice pulling from real client results.

Tours the new backend: a Kanban-style content pipeline where Claude has already planned multiple articles, with the live published piece visible and editable.

Explains the mechanics: a built-in content-writing skill inside the backend repo lets Claude's API write articles in brand voice and auto-publish to the front end; only an Anthropic key and an OpenAI key (for images) were needed manually.

Shows a drafted article moving through review and appearing live and fully formatted on the front end, powered by the shared Supabase database between front and back end.

Recaps the fully working autonomous CMS and teases the next video: building a custom CRM (leads, emails) into the same backend, designed for AI agents rather than clunky legacy CRM tools.
Pairing a strategist model with an execution model, and handing it a concrete architecture spec instead of open-ended instructions, turned a vague 'build me a CMS' goal into a working autonomous publishing system in one sitting.
“Your entire autonomous publishing loop is verified and live end to end.”
“I literally didn't touch anything, and it's actually already gone and published an article live on our page following our exact brand voice.”
“Pretty websites are completely commodified right now.”
“I'm burning through Fable credits like it ain't nobody's business.”
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.
A rebuilt front end with nowhere to publish is just a pretty dead end. This video picks up right where that problem was left: a designer hands his AI coding assistant a single, context-loaded prompt and a pre-written backend spec, then watches it orchestrate a second AI model into building an entire CMS, database, and autonomous publishing loop before he's touched a line of code.
A publicly shared skill (referenced via a GitHub link in the video) formalizes using one AI model as a strategist/advisor and a second as the execution workhorse, so the pricier or usage-capped model is only invoked for judgment calls.
A four-part architecture the creator uses to frame 'owning your stack': a public site and an operating-system-style backend, tied together by one shared database, hosted on infrastructure the business controls.
“the school community is the best place for you to start... learning how to build this system back end, front end is going to increase your value in the marketplace”
Soft pitch woven directly into the technical value section rather than a hard sell — positions the paid Skool community as where to learn the reproducible process behind what's being demoed.
00:00
00:09
00:25
00:32
00:42
00:50
01:01
01:11
01:18
01:30
01:40
01:49
01:59
02:08
02:18
02:27
02:37
02:46
02:56
03:05
03:16
03:25
03:34
03:42
03:53
04:02
04:10
04:21
04:31
04:40
04:50
05:00
05:09
05:19
05:28
05:38
05:47
05:53
06:02
06:20
06:25
06:35
06:45
06:51
07:03
07:13
07:23
07:32
07:41
07:54
07:56
08:10
08:20
08:29
08:39
08:48
08:58
09:07
09:17
09:26
09:36
09:45
09:55
10:04
10:17
10:23
10:34
10:43
10:52
11:01
11:11
11:16
11:30
11:40
11:49
12:00
12:08
12:18
12:27
12:37A six-level stress test of Claude Fable 5 as a web design engine — from a two-minute landing page to a scroll-driven 3D experience that used to take hours.
June 11thAn 84-minute, screen-share walkthrough that takes a v0 template from an empty folder to a live, custom-domain site indexed on Google, Bing, and Yandex — entirely through plain-English prompts.
June 18thHow xAI’s self-describing Agent Instructions let a vibe-coder ship a page-navigating website voice assistant in a single Replit session.
June 17thElie Steinbock walks Greg Isenberg through the build-verify-learn loop he's using to run SEO, ads, and product feedback on autopilot.
July 13thRiley Brown and Ras Mic dig into GPT-5.6, Codex's background computer-use, and why self-scoring agent loops are turning coding tools into a general operating system.
July 12thA creator hands an AI coding agent full tool access, one long brief, and no further prompts — and watches it invent, photograph, animate, and publish more than eighty fictional products end to end.
July 12th