Video summary

You Will Lose Access To Fable Tomorrow. Do This URGENTLY!

Main summary

Key takeaways

Technology

Summary of the Subtitles (Tech Concepts, Product Features, Guides, and Analysis)

Claude Fable Removal Announcement / Timing

  • Anthropic (creators of Claude) is expected to pull the “Claude Fable” model from paid subscriptions around July 7th (possibly early July 8th in Europe).
  • The exact duration is uncertain, but the speaker’s best guess is “a few weeks.”
  • Anthropic reportedly plans to restore Fable to normal subscription availability when compute capacity allows.

Reason for the Change (Cost / Compute Constraints)

  • Fable is described as extremely expensive and compute-intensive, making it difficult to include for all subscribers.
  • The model is switching to a “cost per credit” approach, meaning users must pay based on usage to access Fable.

How to Manage Credits in the Claude App/Site (Feature + Caution)

  • After Fable is removed, the speaker suggests checking:
    • Claude Settings → Usage → Usage credits
  • There is an option to auto top-up credits, which “tops up” after you hit your plan limit.
  • Warning: users must actively monitor usage limits, or they could burn tokens (cost overruns), especially with heavy Fable prompting.

Main “Hack”: Extract Fable into a “Handover Document” Once

The core tutorial is an “extraction loop”:

  1. Use Claude Fable once to generate a complete “handover document” that captures how Fable thinks.
  2. Run that document through a replacement model (e.g., Opus 4.8 or Sonnet) to update/improve workflows.
  3. Reuse the upgraded workflow so you don’t spend Fable tokens on every prompt.

The speaker claims this reduces reliance on Fable execution and turns Fable’s “brain” into something that can be translated into skills/project instructions.

Key Prompt Characteristics (What the Handover Document Must Include)

The described prompt instructs Fable to write like a retiring senior analyst brain dump for its successor. It should include:

  • Way of thinking (not a rules checklist)
  • Problem decomposition
  • How to interpret intent
  • Verification / self-review (avoid only pattern-matching)
  • How to communicate conclusions

Concrete Workflow Integration

The speaker demonstrates applying the extracted document to:

  • A Claude “skill” (example: a video production pipeline that improves planning via a feedback loop)
  • A project folder instruction set, so the document influences future chats within that project

They also recommend using skills for training loops, rather than repeatedly prompting Fable directly.

Skill-Building Workflow (Tooling Mentioned)

  • Skills can be created using something like / / skill creator (slash command).
  • The speaker mentions using WhisperFlow for voice transcription, enabling interactive skill creation via spoken responses.

Limitations Acknowledged

  • This is not a way to get “exact Fable intelligence” on every prompt.
  • Because Opus/Sonnet cannot fully “execute Fable-like behavior” the same way, the document mainly improves:
    • Strategic quality
    • Outputs (especially for planning/analysis)
  • Nuance may still differ for cases like certain code/brainstorming edge cases.

Cost Analysis (Credits / API) + Recommended Usage Strategy

  • Fable is framed as one of the most expensive models.
  • Costs described:
    • Extraction / handover: “a few dollars” (few credits)
    • Large sessions (e.g., coding): potentially $10–$20 per session input and ~$50 for output (as stated)
  • Execution recommendation:
    • Don’t do heavy execution on Fable.
    • Use Sonnet/Opus as executors and use Fable mainly for strategy/advice.

“10–80–10” Strategy (Cost-Effective Orchestration Concept)

  • Use Fable for:
    • 10%: orchestration / planning / review (strategy)
    • 80%: execution by cheaper models (Sonnet or Opus)
    • 10%: review / tweaks by Fable
  • Claimed outcome: daily-driving Fable costs drop from very high amounts (if used constantly) down to ~$3–$10/day.

Best Alternative After Cut-Off: “Advisor Tool” Agent Setup

The speaker’s “real source” strategy is presented as an implementation using an advisor tool (linked via platform docs). Conceptually:

  • Fable acts as an on-demand strategist/advisor
  • A system updates context (via Claude memory or MD files in a folder/history)
  • Sonnet executes loops (example: via /loop)
  • Fable is invoked only when advice is needed (beginning/end of the process or specified checkpoints)

The speaker claims this avoids wasting tokens because Fable is slow at execution, while Sonnet/Opus are faster.

Developer / API Access Mention

  • Power users can use platform.claude.com/dashboard to get an API key to access Fable via the API route.
  • The speaker suggests there’s confusion around API vs credit pricing and notes they’ll cover costs later (or in the document).

Main Speakers / Sources

  • Primary source / speaker: The video narrator (single person) explaining the Claude Fable changes and providing the tutorial/prompt/system.
  • Referenced organization: Anthropic (Claude creators), including “official developers” citing timing/capacity rationale.
  • Referenced tools/products: Claude app/website settings, Claude API, platform.claude.com, WhisperFlow, and Claude skills / agent tool documentation (including the advisor tool).

Original video