← Claude
David Ondrej
Fable 5 is back… here is my plan
⏱ 31 min video · 3 min read2 Jul 2026Worth watching
TL;DR
David Ondrej discusses the return of Claude 5 (referred to as 'Fable 5' throughout) after its brief ban, sharing his workflow for maximizing the model before potential API-only restrictions, strategies for using it alongside cheaper open-source models, and his long-term plan to build a self-hosted GPU cluster to avoid dependence on cloud AI providers.
Key points
1
Claude 5 was banned after an Amazon employee, contractually obligated to the US government, reported a security vulnerability discovered during routine code-fixing prompts — not a deliberate jailbreak attempt.
2
David recommends using Claude 5 as the orchestrator/planner while routing execution tasks to cheaper open-source models like Kimi K2, GLM 5, and Minimax — cutting costs by up to 25x.
3
He created a free 'FableSafePrompt' skill that rewrites prompts to avoid triggering Claude 5's classifier, reducing downgrades to Opus without changing the intended output.
4
He is investing $50,000-$100,000 over six months to build a private GPU cluster in Katowice, downloading model weights and Hugging Face datasets to a NAS expanded to 80TB+ for full local sovereignty.
5
His startup thesis: all software will primarily be used by agents, not humans, so builders must prioritize clean APIs, CLI tools, and agent-usable interfaces over traditional web UIs and SaaS subscription models.
Actionable insights
→
While Claude 5 is available, use it to generate long-term plans for fitness, business, health, and other life areas, and to write SOPs/skills so weaker models can follow superior instructions later.
→
Set up automated scripts (on a VPS) to pre-populate datasets and answers from Claude 5 now, so you retain that intelligence even if the model becomes unavailable again.
→
Delete 80-100% of your existing AI skills/prompts and rebuild only what Claude 5 genuinely cannot do by default — the model is capable enough that most handholding is now obsolete.
→
Build or architect any software product to be agent-usable first: clean API, good documentation, CLI support — because agents will soon represent the majority of software consumption.
→
Download the free FableSafePrompt skill (linked in video description) to rewrite borderline prompts and stay on Claude 5 instead of being downgraded to Opus.
Notable quotes
“I would trust Fable 5 to run the European Union way more than the current leaders.”
“If you don't find yourself adding back at least 10% of the things, you are not deleting enough.”
“In the future, intelligence is going to be the most important resource on earth.”
Worth watching?
✅
Worth watching the full video?
Worth watching if you actively use Claude in coding workflows and want the multi-agent cost-cutting strategy and free prompt skill explained in context — but the core value is fully captured here.
Saved you some time? The creator still deserves a like.
Want this for your own channels?
Add the channels you follow. Every new video summarised and in your inbox the moment it drops. From £4/month.
Try it free