How I wrote with AI the book that Amanda Askell never wrote

2026-06-06 · Source: Génération IA · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Advanced, long

Summary

The article details the creation of "Murmures - Amanda A.", a book synthesized from Amanda Askell's scattered interviews and writings using artificial intelligence. Authors Benoît Raphaël, Thomas Mahier, and Jeff GPT initially attempted a single-prompt approach, which quickly led to degraded thought and inconsistencies. They subsequently developed a refined method, entrusting Fable 5 (Claude) with orchestrating source collection, analysis, and organization, while GPT-5.6 Sol (Open AI) handled text generation using a specialized "skill" and "style recipe." The process involved multiple AI proofreading agents for content validation, linguistic quality, and voice consistency, alongside extensive human review. This approach, though taking several weekends, significantly reduced the time compared to traditional methods, emphasizing the importance of human oversight and "functional empathy" in complex AI-driven projects.

Key takeaway

For AI Engineers developing complex content generation systems, this project demonstrates that chaining multiple AI operations requires a highly structured, multi-agent approach with significant human intervention. You should design workflows that distribute tasks among specialized AI agents and integrate robust human and AI-driven proofreading to ensure fidelity and coherence, especially when synthesizing existing intellectual property. This iterative process fosters deeper understanding of both the subject and the AI's capabilities.

Key insights

Complex AI writing projects demand structured multi-agent workflows and human oversight to maintain coherence and fidelity.

Principles

• AI struggles chaining novel, complex operations.
• Preserve author's spirit, not literal text.
• Functional empathy improves AI instruction clarity.

Method

Orchestrate source collection and analysis with one AI (Fable 5), then use another (GPT-5.6 Sol) with a "skill" for writing, supported by multiple specialized AI proofreaders and human review.

In practice

• Distribute complex tasks among specialized AI agents.
• Implement AI proofreading for voice and accuracy.
• Use the provided prompt for creative concept illustration.

Topics

• AI Content Generation
• Multi-Agent Systems
• Large Language Models
• Prompt Engineering
• AI Workflow
• Text Synthesis
• Functional Empathy

Best for: AI Engineer, Prompt Engineer, Director of AI/ML

Related on AIssential

• Writing in the Age of AI 2.0: Progress Sets the Direction, but Does Not Guarantee the Result — AI in writing should automate routine tasks and augment thinking, not replace it, with human oversight crucial.
• I Built a Fully Automated AI Publication in a Few Hours. Here’s Exactly How It Works. — Transparent AI-generated content, with distinct bot personas, can create engaging and cost-effective digital publications.
• Why Every AI Writing Tool I Tried Failed Me (And What They All Get Wrong) — AI writing tools fail at novel-length consistency due to context amnesia and a lack of structured, persistent world-building.
• The Truth About AI Writing — The suitability of AI writing depends on content purpose and reader expectations, not just authorship.
• My Make-Believe Project Hail Mary: Writing with Generative AI as Unknown Intelligence — Simulating "unknown intelligence" with generative AI reveals effective communication strategies.
• Improving Collaborative Storytelling with a Multi-Agent Framework Based on Large Language Models — An iterative LLM Writer-Editor framework significantly improves collaborative storytelling quality, even with few refinement steps.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Génération IA.