Building Real Apps with Claude Fable

The Breakthrough

Claude Fable demonstrates a significant leap in reasoning capability that allows it to execute complex, multi-file application expansions in a single pass, provided the user defines explicit "definition of done" verification criteria.

What Actually Worked

• Strategic Shaping: Before coding, the author used Claude as a thought partner to define scope, database entities, and technical requirements, resulting in a comprehensive scoping document rather than a simple feature request.
• Verification Criteria: The author included a specific "definition of done" checklist within the prompt. This allowed the model to self-test its output against predefined requirements, reducing the need for manual "no, not like that" feedback loops.
• Agentic Workflow: The author utilized a "night shift" pattern where the application (a Rails-based tool called Residents Radar) serves as the UI/API layer, while AI agents run on recurring schedules to perform data extraction and analysis.
• Clarifying Questions: By explicitly instructing the model to ask clarifying questions before starting, the author forced the model to perform a deep-dive analysis of the existing codebase, which surfaced missing infrastructure details like specific background job configurations.

Before / After

• Refinement Effort: Previously, the author's workflow required multiple rounds of back-and-forth refinement to fix implementation errors. With Claude Fable, the refinement stage is significantly reduced because the model correctly interprets complex instructions on the first attempt.
• Model Cost: Claude Fable is significantly more expensive than Claude 3 Opus, with the author noting it is roughly twice as costly in terms of API token consumption.

Context

Building custom tools is now accessible to non-technical users, but the human role has shifted from "coding" to "shaping." The author emphasizes that the quality of the output is directly proportional to the quality of the initial planning and the clarity of the verification criteria provided to the model. As models like Fable become more capable, the primary developer skill is no longer just writing code, but deciding which tasks warrant the high cost of a "heavy hitter" model versus a more economical daily driver.

Notable Quotes

• "The refinement stage is starting to melt away... that stuff is reducing since Fable seems to be really good at checking its own work as long as you give it clear notes on what done looks like."
• "Choosing the right model is now the new skill because this capability is expensive."

Content References

• { "type": "tool", "title": "Residents Radar", "context": "mentioned" }
• { "type": "tool", "title": "Claude Code", "context": "mentioned" }
• { "type": "tool", "title": "PRD Creator", "url": "https://buildermethods.com/prdcreator", "context": "recommended" }