I recently stumbled upon DTV, the book that Sir Michael Moritz, historical partner at Sequoia, wrote to remember and celebrate Donald Thomas Valentine, the iconic founder of Sequoia.
Apparently there are only a few printed copies in existence, some curiously available at an Italian hotel in Tuscany owned by Moritz. This post is about the book, and how I used AI to bring it back to life as an ebook that you can download below.
As I get to spend more and more time in venture helping companies at Axiom, I find myself thinking a lot about what real venture is and what made the great venture firms great.
Today many investors are chasing what’s obvious in AI to generate quick markups. This is not venture.
At Axiom, we’re aiming to bring back what venture used to be. We back founders who have a different view of what tomorrow looks like and are using AI to build the next generation of category-defining companies. We roll up our sleeves to help them build and scale. We have the patience to create something that compounds for decades and changes people lives.
DTV seemed like the perfect source to understand what venture used to be. And what it should be again.
Reading Valentine’s approach brought back so many lessons: his obsession with large markets, high gross margins, and low burn rates. His Socratic questioning - ”Why?” “Who cares?” “Who needs it?” “What does it do?” His belief that understanding technology deeply was how he “could see the future.” His conviction that “it’s disobedience that gets you innovation.” This clarity of thinking, this discipline, this focus on what actually matters - that’s what we’re trying to bring back.
The book also brought back memories of Cisco, Don Valentine’s most successful investment in those early days of Sequoia, and where I worked early in my career. The section on Sequoia’s investment in Crescendo and its acquisition by Cisco struck a chord, as I had the honor to work with Crescendo’s founders in my first startup. From them I learned what it actually means to build a startup: the obsessive focus on product excellence, the relentless drive that doesn’t accept “good enough,” the culture of ownership where everyone acts like a founder, the mindset that speed and execution matter more than anything else. Don’s perspective on that acquisition validated everything those founders taught me — lessons I want to pass on to the entrepreneurs we work with today.
How I Recreated the eBook with AI — Without the Book
From that hotel in Tuscany, I began my online hunt for the book. The Internet did its magic and published the pictures of the entire book, first on Google Drive and then on Slideshare.
But in the age of AI, I couldn’t bring myself to read pictures of distorted pages:
So I decided to start a little project to test, as I do on a daily basis, the capabilities of AI models.
First I uploaded the whole 53.7MB PDF to ChatGPT. Got an error. Tried just a few pages, still got an error:
I knew I couldn’t count on Claude to handle a file of this size, but had to try:
I then tried Grok, which is always pretty resourceful in these extreme cases. The answer was actually funny:
Last hope was Gemini, which I access mostly via AI Studio; I selected the small but mighty Gemini 2.5 Flash, as I didn’t need the advanced reasoning from Pro. Somewhat hopeless, I first uploaded a screenshot of two pages to see if at least that would work before spending more time. It did! I just had to refine the prompt a few times to get the formatting I wanted. Once the result was satisfactory, I asked Gemini to craft a polished version of the prompt, which I then set as the system prompt.
From there I started simply copying and pasting blocks of 5 pages each from the PDF into AI Studio. It worked beautifully, so after a couple of uploads I started increasing to 10-page batches and it still worked. Very quickly I was done transcribing the entire book. Nice!
Of course now I didn’t want to just read a text file. So I decided to ask Claude how to create an ebook, something I’ve never done before. In the age of AI, there is no excuse not to do things. I asked Claude as I assumed there was some coding involved, and in fact it did.
The code, short and simple, worked great right away, but I had to face another limitation: Claude artifacts couldn’t handle the length of the content I shared and only converted the first few pages of the book to HTML, the format needed to generate the ebook.
So I switched to Windsurf to write a script to convert the plain text of the book to HTML, to be added to the code written by Claude. The HTML version of the book now looked pretty good; I just spent some time tweaking the code to more closely follow the paragraph formatting.
Solved this problem, it was time to generate the actual ebook. I reverted back to ChatGPT to figure out the next steps. An open source app, Calibre, was recommended as the most flexible and advanced option. After installing the app, I found myself trying to figure out the somewhat overwhelming UI. As I often do these days, I didn’t even try to read the suggested manual or watch the recommended YouTube video and just asked ChatGPT what to do.
In a few steps I was able to generate the first version of the ebook, getting familiar with Calibre’s UI and its structure. The main issues were the first two pages: the cover and the internal title page.
The cover was some pretty ugly template from Calibre, which I promptly decided to replace with the actual one posted from the hotel in Tuscany:
Considering we are in the age of AI, I decided to use Nano Banana from Google to remove the background. Surprisingly, I had to go through quite some back and forth, because the image was simply returned as is. Finally, I was able to get the cover on top of a white background that I then cropped and very slightly rotated with the Photo app on my iPhone.
At this point, I wanted to also reproduce as accurately as possible the title page:
I took a screenshot of the signature and, because we are in the age of AI, I tried to remove the background and straighten the signature using again Nano Banana.
The result was just bad. Despite several attempts, guiding lines to help the model, and a task clearly understood, the model couldn’t complete what I thought was a relatively simple task:
A task that I easily completed using again the Photo app on my iPhone.
I then stubbornly asked the model to remove the background from the image I provided. The result was something that looked like it had a transparent background, but it then became clear the model was just mimicking transparency by drawing the checkered pattern itself (those gray and white squares that image editors use to represent transparent areas) directly into the image rather than actually removing the background:
I turned back to an old fashioned tool, Preview, and obtained the signature with no actual background.
As you can tell, at this stage of AI you need not only agency, but also persistence and the ability to pull the right tools together. But ultimately, a book that was impossible to get is now available:
This is awesome, thank you
Wow! Seriously, thank you!!