In February 2026, I sat down on a Saturday morning with a cup of coffee, a detailed outline, and the early version of what would become InkEngine. By Sunday evening, I had a 56,000-word manuscript for The Autonomous Engineer - a book about network automation that I'd been meaning to write for three years.

Three years of "I should write that book someday." Two days of actually doing it.

I want to walk through exactly how this happened, because the headline is misleading in an important way. I didn't sit down cold on Saturday morning. The two days of writing were preceded by weeks of preparation that made the fast generation possible. Speed without preparation produces garbage. Speed with preparation produces a real book.

The Preparation (2 Weeks Before)

The writing happened in two days, but the groundwork started earlier. Here's what the preparation phase looked like:

Week 1: Interview and voice profiling. I spent about 4 hours over three sessions doing the InkEngine interview. I talked about my 15 years in network engineering, the shift from manual CLI configuration to infrastructure-as-code, and why I thought most automation books missed the point. I also fed the system about 12,000 words of my existing writing - blog posts, conference talk scripts, and internal documentation I'd written.

Week 2: Outline and chapter planning. The expert panel - five AI personas representing different reader perspectives - helped me brainstorm and structure the book. I came in with a rough 8-chapter outline. We ended up with 14 chapters after the panel identified gaps and suggested splitting some topics that were too dense for single chapters.

Each chapter got a detailed brief: the main argument, key examples I wanted to include, target word count, and notes about how it connected to surrounding chapters. This took about 6 hours spread across the week.

Total preparation time: roughly 10 hours. That's the hidden cost behind the "2 day" headline. But 10 hours of planning plus 2 days of generation is still dramatically faster than the 6-12 months most non-fiction books take.

Day 1: Saturday

7:00 AM - Started generation. I kicked off Chapter 1 and watched the first draft come together in about 4 minutes. Read through it while drinking coffee. The voice was close but the opening was weak - too much setup, not enough hook. I adjusted the chapter brief and regenerated just the first section. Better.

7:30 AM - 12:00 PM - Chapters 1-7. I got into a rhythm. Generate a chapter (3-5 minutes). Read through it while the review panel ran (60-90 seconds). Check the scores. Review any flagged sections. Make notes for edits I wanted to do later. Move to the next chapter.

The review panel caught problems I would have missed on a casual read-through. Chapter 3 had a Pattern Scanner score of 5 - too many AI tells. The system revised it automatically and the second version scored 8. Chapter 5's Voice Matcher flagged that my sentence length was running too long compared to my profile. After revision, it tightened up.

By noon I had 7 chapters totaling about 28,000 words. I took a break.

1:30 PM - 3:00 PM - Chapters 8-10. The afternoon chapters were technically denser - covering specific tools like Ansible, Terraform, and Nornir. These needed more from my interview data because they relied heavily on my actual experience with these tools, not generic descriptions. The system pulled from my interview stories about migrating 2,400 switches and the time a misconfigured Ansible playbook took down a data center link.

3:00 PM - 5:00 PM - First editing pass on chapters 1-7. I went back to the morning's work with fresh eyes. Fixed some awkward transitions. Added a personal story to Chapter 2 that the AI didn't have (it happened after my interview). Removed a section in Chapter 4 that was accurate but boring. Rewrote the conclusion of Chapter 6 because it summarized too much - I wanted it to point forward to the next chapter instead.

End of Day 1: 10 chapters drafted (approximately 41,000 words). About 9.5 hours of actual work with breaks.

Day 2: Sunday

8:00 AM - 11:00 AM - Chapters 11-14. The last four chapters covered advanced topics: CI/CD pipelines for network changes, testing frameworks, observability, and a forward-looking chapter on where network automation is heading. These were the chapters I was most worried about because they required the most specific technical knowledge.

The system handled them well. My interview had covered these topics in depth, and the chapter briefs were detailed. Chapter 13 (observability) needed the most manual editing because the space has changed fast and I wanted to reference some recent developments the AI didn't know about.

11:00 AM - 2:00 PM - Editing pass on chapters 8-14. Same process as Saturday afternoon. Read through, fix problems, add missing details, cut what didn't work. Chapter 12 got the heaviest editing - I essentially rewrote two sections because my thinking on testing had evolved since the interview.

2:30 PM - 5:00 PM - Full read-through. I read the entire manuscript from Chapter 1 to Chapter 14. This is where you catch continuity issues - a concept explained in Chapter 3 that's re-explained in Chapter 7 as if the reader hasn't seen it. A promise made in the introduction that never gets fulfilled. Tone shifts between chapters written at different energy levels.

I found about 30 things to fix in the full read. Mostly small - a repeated example, an inconsistent term, a chapter ending that didn't transition well to the next. I spent the last 90 minutes making these fixes.

End of Day 2: Complete 56,000-word manuscript. 14 chapters. About 8.5 hours of actual work.

The Numbers

• Total preparation time: ~10 hours (interview + outline)
• Total generation + editing time: ~18 hours across 2 days
• Grand total: ~28 hours from first interview to complete manuscript
• Final word count: 56,247 words
• Chapters: 14
• Average chapter length: 4,018 words
• Review panel average score: 8.1/10 across all reviewers
• Chapters requiring revision loops: 9 of 14 (64%)
• Maximum revision loops for any chapter: 2

What Happened After

The two-day manuscript wasn't the final product. After the writing weekend, I spent another week on:

• Fact verification: The Fact Checker reviewer had flagged 47 claims across the book. I verified each one. Three were wrong - AI hallucinations about specific version numbers and release dates. Easy fixes, but they would have been embarrassing in print.
• Professional editing: I hired a technical editor who specializes in DevOps/networking books. She spent two weeks with the manuscript and came back with structural suggestions for three chapters and about 200 line-level edits. Cost: $1,800.
• Beta readers: Four colleagues from the network automation community read the manuscript. Their feedback was overwhelmingly positive on content and voice. Two of them asked how I'd found time to write a book while running InkEngine. They didn't guess AI was involved until I told them.

Would I Do It Again?

Absolutely. I'm already planning my next book. The process isn't magic - it requires real expertise and real preparation. You can't interview your way through a topic you don't understand. The AI amplifies what you bring to the table; it doesn't create knowledge from nothing.

But if you have the expertise and you've been putting off writing because the blank page is too intimidating or you don't have 500 hours to spare - this changes the math completely. Twenty-eight hours to go from "I should write a book" to a complete, reviewable manuscript. That's not science fiction. That's what I did three weeks ago.

Lessons Learned

Preparation quality determines output quality. The detailed chapter briefs were the single biggest factor in the manuscript's quality. Chapters with thorough briefs needed minimal editing. Chapters where I'd been lazy with the brief needed heavy revision.

Don't skip the editing pass. It's tempting to generate all chapters and call it done. Don't. The editing pass is where you add the personal touches, fix the things only you would notice, and turn a good draft into a great one.

The review panel earns its keep. Without the automated review, I would have published text with obvious AI patterns. The Pattern Scanner caught things I was blind to, especially after hours of reading AI-generated text when my calibration was shot.

Tell people about your process. I was nervous about disclosing AI involvement. Turns out, most readers don't care how you wrote it - they care whether it's good. And the people who do care are more impressed than offended when they learn about the pipeline.