AI Writing Process¶
How AI was used to write an 81,000-word book with 775 citations, consistent voice, and multi-layer editorial review.
This isn't a blog post about prompting ChatGPT and calling it a book. This is the documented system behind Blueprint for An AI-First Company -- 12 chapters, 775 inline citations with source URLs, and a voice that reads like one human wrote it. Every prompt, script, skill, and architectural decision is here for you to study, adapt, or steal.
What This Covers¶
The full AI-assisted book production system, end to end:
- Author voice encoding -- 6 files that teach an LLM to write like a specific human, not like an AI
- Prompt engineering -- 27 modular prompts across 5 categories (writing, editing, review, research, analysis)
- Multi-agent orchestration -- 14 Claude Code skills that act as writer, reviewer, researcher, and quality auditor
- Research pipeline -- 180+ Perplexity prompts that generate citation-ready research with source URLs
- Obsidian vault architecture -- The file structure, templates, and linking system that kept 12 chapters organized
- Automation -- 17 Python scripts for PDF generation, word count analytics, citation management, and vault health
- Book intelligence app -- A Flask + PostgreSQL pipeline with 70+ modules for cross-chapter analysis
- Editorial review -- A 4-phase process covering structural editing, voice consistency, fact verification, and contradiction detection
By the Numbers¶
| Metric | Count |
|---|---|
| Total words | 81,000+ across 12 chapters |
| Inline citations | 775 with source URLs |
| Writing prompts | 27 modular prompts in 5 categories |
| Claude Code skills | 14 (writer, reviewer, researcher, quality auditors) |
| Research prompts | 180+ for Perplexity automation |
| Python scripts | 17 for automation and analytics |
| Voice system files | 6 defining author tone, style, and guardrails |
| Editorial phases | 4-phase review process |
| Review dimensions | 10-dimension acquisition-level quality audit |
| Intelligence app modules | 70+ in the analysis pipeline |
| Documentation files | 57 across 12 directories |
| Documentation words | 51,000+ with 7 Mermaid diagrams |
| Adaptable templates | 17 fill-in-the-blank starting points |
Navigation¶
| Section | What You'll Learn |
|---|---|
| 01 -- Overview | End-to-end pipeline, architecture decisions, and results |
| 02 -- Author Voice | Building a voice system, gold standard method, drift prevention |
| 03 -- Prompt Engineering | 27 prompts, the master system prompt, writing and editing prompts |
| 04 -- Agent System | Multi-agent architecture, chapter writer workflow, quality skills |
| 05 -- Research Pipeline | 180+ prompts, Perplexity automation, citation management |
| 06 -- Obsidian Vault | Vault architecture, templates, linking conventions, dashboards |
| 07 -- Automation | 17 scripts for PDF generation, analytics, and vault health |
| 08 -- Book Intelligence App | Flask + PostgreSQL analysis pipeline, 70+ modules |
| 09 -- Review Process | Multi-layer review, editorial workflow, contradiction detection |
| 10 -- Lessons Learned | What worked, what failed, what I'd do differently |
| Templates | Adaptable starting points for voice, prompts, skills, vault, and scripts |
Who This Is For¶
- Non-fiction authors who want to use AI as a writing partner without losing their voice
- Technical writers building documentation systems that need consistency at scale
- Content teams exploring multi-agent workflows for long-form production
- Anyone curious about what AI-assisted creative work actually looks like when you take it seriously
How to Use This¶
Start with 01 -- Overview for the full picture -- how the pieces connect, why certain architecture decisions were made, and what the results looked like.
Jump to any section that matches your immediate need. Each one is self-contained with enough context to be useful on its own.
Grab from Templates when you're ready to build. These are fill-in-the-blank starting points adapted from the actual system -- voice profiles, prompt structures, skill definitions, vault layouts, and automation scripts.
Tech Stack¶
| Tool | Role |
|---|---|
| Claude Code (Opus 4.5) | Writing, editing, review, agent orchestration |
| Perplexity Pro | Research and citation generation |
| Obsidian | Vault-based manuscript management |
| Python | Automation scripts, analytics, PDF generation |
| Flask + PostgreSQL | Book intelligence app and cross-chapter analysis |
| Playwright | Browser automation for research workflows |
About the Book¶
Blueprint for An AI-First Company covers how companies can build AI into their core operating model -- not as an add-on, but as the foundation. 12 chapters across four parts: Foundations, Building, Operating, and Sustaining. The process documented here is itself a case study in AI-first production.
Built by Saurav as part of writing Blueprint for An AI-First Company.