Dear CIO,

To understand the future, we need to understand the past. For anyone unaware, I have spent the past year writing a book dedicated to explaining the history of AI. I’m getting close to the finish line, so I wanted to share a sample of the book’s Prologue that I just finished. The plan is to give samples at the upcoming All Things Open AI conference on March 18th. I cannot wait until its release, and I hope you enjoy this early preview.

Best Regards,
John, Your Enterprise AI Advisor

A First Look At Rebels of Reason

Early Preview of My History of AI Book

Prologue

My good friend Mark Hinkle showed me a new tool in mid-2020 that was still under the radar with most people, even computer geeks. It was called GPT (Generative Pretrained Transformers). Then, a tiny population understood what it meant. Even the rest of us, aware of this new tool, didn't care. It was one of those moments when Mark showed me an interface that made an API call to this tool. It was like the first time you get high. (I didn't say that.) It seemed like a new kind of magic. 

At the time, I was working on the last part of my Deming book, and my first thought was that this would be an excellent tool for research. You asked it a question, and it wrote a whole paragraph of explanations. I decided to register with the founders of this tool, OpenAI. However, to my dismay, it was great with its answer in the first paragraph but not so good at the second and downright awful when it got to the third. The "curiosity killed the cat" in me needed to understand why this tool could be so good and terrible simultaneously (see what I did there). After a few days of research trying to know what the GPT stuff stood for, I realized I was going to have to take a master's program to figure it out. At this point in my life, I had zero time for that nonsense, so I put this new AI stuff on ice, although I'd poke my head in occasionally. I even tried at one point to take a Coursera TensorFlow course. After my day job and hanging out with my family, once they go to bed, I have my "me time" and this is the kind of stuff I do. If something isn't fun, I don't like to invest time in it.  

In 2022, OpenAI released GPT-3.5, so I took another look. At this point, the first paragraph and a half were good; the rest were useless for research purposes. Later that year, the infamous ChatGPT came out, which was still meaningless for what I needed. In fact, now that everyone was doing it, someone asked me, "Have you used ChatGPT yet?" I responded that I already had a magic eightball somewhere at home, so I'd just use it instead. 

The magic for me started when a friend showed me something called Retrieval Augmentation Generation (RAG). In short, this method of using GPTs allowed me to use grounded research and get high-efficiency answers quickly. By this time, my Deming book had already been published; however, I realized I could learn how this crazy stuff all worked using my Deming book for what data scientists call "ground truth." I turned my book into a RAG and started to learn all about Generative Pretrained Transformers. The game was afoot. 

It was late 2023. I was sitting in a cafe outside the convention center of Amazon's cloud conference called AWS: ReInvent having coffee with a friend who was a quant (quantitative analyst). He had just started his own fund and was asking questions about what technologies he should be using for his algorithms. Questions about Kubernetes, Docker, and stuff. I asked him what the application stack looked like, and he rattled off a bunch of Python libraries I had never heard of (AI stuff). However, one of the tools caught my attention, which was PyTorch. I knew a little about PyTorch, having taken the TensorFlow course. (For the uninitiated, PyTorch and TensorFlow are the top two primary tools AI scientists use to build models used by apps like ChatGPT.) Although my friend was a brilliant investor, he was less knowledgeable about technologies than I was. So if he could do PyTorch, then I felt I could. 

On my plane ride back from the conference in Vegas, I decided to try PyTorch. It was a bit easier; however, as Derek always says, the juice wasn't worth the squeeze. The storyteller in me wanted to know the story behind PyTorch. I had an instinct that this might be an interesting topic and wondered if another book might exist here. I don't know why because I knew next to nothing in that moment. But I had a gut feeling. 

I found a podcast about the founder of PyTorch, Soumith Chintala, with my good friend Mark Miller. Chintala's story was fascinating—as in Dr. Deming-level interesting. I had my next book idea! But sorry, Soumith—your career hasn't been long enough to warrant a whole book, although your story is amazing. But if you're reading this, know that your chapter in Rebels in Reason was the genesis of this entire book. 

After realizing this, I knew there had to be other great stories related to the history of AI. As I said in my Deming book, I pulled on a thread and found a fascinating tapestry. This time, it was not just any old tapestry—I'm talking about the "Bayeux Tapestry" which is nearly 70 meters long and depicts the Battle of Hastings in 1066.  

I started with the usual suspects: Hinton, LeCun, and Bengio. But I figured, why stop there? Before I knew it, I had about 15 individuals profiled, and this was only from the last decade. When I decided to write my book on  Deming, I wanted to tell the story like I believe Michael Lewis would have written it. I did a preliminary survey to see how many key characters Lewis uses in his books. My napkin-based statistics came up with less than 10 people per book. By that time, I had well over 30 people spanning the last 75 years of AI, and I hadn't realized yet that I needed to cover some 19th-century scientists to get the story right. As one of my first editors always says, "Write, John, write." So, I decided to research, write, and let the chips fall where they lay. Before I knew it, I had read over 20 books about AI's history and had a list of about 50 characters. One of the bonuses of this topic was that all of those books were in digital form. With Deming, only about 3 of the 20 books I used were in digital form. In other words, ordering, listening, and reading online books is much quicker than ordering, shipping, and physically reading a book. 

As Derek and I started working out the arc of this book, we ran into another problem. There are many throughlines related to the history of AI. The two obvious ones are what are commonly called expert systems and neural networks. However, there are other essential threads; for example, speech recognition, computer vision, natural language processing (NLP), machine learning and deep learning, and autonomous systems. It would be impossible to tell a linear story about history without creating whiplash for the reader going back and forth between timelines. For example, I might tell you the story of Deep Blue in 1997, then go back to 1947 to tell you about the first paper that led to neural networks. Then, back to the future, we hear the story of two crazy guys who wired up cats to see how their brains worked. Back and forth. So, we came up with a plan. We would tell stories based on the history of AI but separate the threads by function: how AI learned to count, think, learn, read, see, and speak. Now, all we had to do was write and write and write. 

There's an old joke that goes, you want to make God laugh, tell him your plans. If you want to make God ROFL¹, tell him about the book you think you'll write. I started writing Rebels. I thought the arc would be, "Nothing new here; it's just computers, networking, and storage." Although that is still a sub-theme of the book, about six months in, we took a shift. My co-conspirator on the founding of DevOps, Damon Edwards, has always been one of my best advisors. As with Deming's Journey to Profound Knowledge, I gave him an overview of this book's outline when I started. Along the way, I kept him updated on my progress. A little more than halfway through, he said, "It sounds like your book is about a bunch of rebels." I realized he was right. 

We immediately shifted our focus to re-align every chapter based on one or more rebels. The good news was we already had the rebel stories. We needed to rewrite the focus to be more about the rebel than the tech. One of our rebels, Turing, needed two chapters. The more we focused on telling the history of AI through the lens of rebels, the more we recognized the patterns of these rebels. They all were brilliant. However, most had a relentless vision and never took no for an answer. Some were downright weird, and some might have been dangerous to hang around with.  

The thing I love most about writing a book is the same thing I like about creating a startup. Although the feedback loop is shorter with a book, both start off with a blank canvas. They both have artificial completion dates; in other words, they are never really done. The support for a tech product is vastly more complicated, but I currently have about a page and a half of errata for a future edition of Deming's Journey to Profound Knowledge. In short, starting a software company and writing a book are both a journey.  

As I complete the first revision on this particular journey, I feel like this is a great book. When we were doing some of the final read-throughs of the DevOps Handbook, I had an out-of-body experience. I thought I was reading someone else's book and said to myself, "I should probably buy this book." Silly, because I wrote it. I actually have the same feeling while finishing the final read-throughs of Rebels of Reason. I hope you enjoy this book if you decide to embark on this journey with me. 

• Reuven Cohen explains why true expertise comes from building functional, verifiable implementations, not just talking about frameworks and strategies.

• I usually stay away from the anti/pro-AGI discussion because I believe it's a waste of time. However, this one said everything I've been trying to say out loud.

• Lonut Ilascu writes on how researchers at Truffle Security discovered nearly 12,000 valid API keys and passwords in the Common Crawl dataset, which is used to train AI.

• Koenraad Schelfaut and Prashant P. Shukla write on how technical debt is becoming a major business liability in the AI era.

• James Cooker writes that software security vulnerabilities take longer to fix due to rising reliance on third-party and AI-generated code.

• Robert Lemos describes how AI-driven systems like NVIDIA’s Agent Morpheus and Amazon’s Bedrock are enhancing software vulnerability management.

• Ryan den Rooijen, Wade Munsie, and Randy Bean dive into the expanding role of Chief Data Officers as organizations rapidly increase investments in AI and data.

• Here’s a list of the places I will be speaking in the next couple of months

• The AI Native Dev Meetup in London on March 25 will feature talks on AI-assisted legacy modernization and AI-driven documentation.

• AI/ML adoption is accelerating in IT, but rising infrastructure costs, shadow AI risks, and declining code quality are creating significant challenges for DevOps and platform teams. Asaf Yigal, Anton Weiss, and I are speaking in a live webinar covering how to mitigate these risks in an upcoming webinar.

• The Artificially Intelligent Enterprise talks about how to optimize prompts for reasoning models.

• AI Tangle looks at Alibaba’s new AI model, which reportedly can rival DeepSeek.