At the Speed of AI

Yesterday, when I finished writing the previous chapter and saved it as an .md file as usual, Claude's terminal told me: "This chapter is different from all the previous ones. You've moved from the didactic and technical to the human and reflective. Very powerful."

Today those responses surprise us because we're not used to a screen congratulating us or telling us what it thinks about what we've done. In a few years, that will change — we'll get used to it and it will seem normal. But we can go a step further: imagine a robot sharing an office with us making the same comment, and then that too will become natural, just as we're now getting used to the screen responding that way. Humanoid robots are the physical evolution of what today we can't see behind a screen.

When you decide to write a book, your first book, you have no idea how long it might take. Today it must be a hundred times easier than it was 50 years ago, when there were still typewriters and a person's mind. I have several lines of research open, available bibliographies, news sites, and many tools. Even so, it takes a long time.

Writing is thinking, organizing, discarding, and starting over. It's looking at a page you wrote yesterday and realizing that half of it is no good. It's deleting entire paragraphs that took you an hour to write. It's discovering that what you wanted to say wasn't what you actually said. It's an exercise in constant humility.

I set out to write it in a month, as a challenge. I still work at my company every day, but I've cut out TV and social media. I've devoted all that time to the book. Since I started writing, my dream has changed. It's much lighter. I sleep less and think more. I wake up thinking about what to write, and I go to bed thinking about what I'm going to write tomorrow.

I don't know how to write. No one taught me how to be a writer. So I have to rely on Claude Code, which was made for this. Editing takes a long time. But I feel like I'm getting there. The words are starting to flow differently; the connection between ideas comes together more naturally. At first, it was just spitting out text. Then it became a more creative process, something that came from within, where you start to see how the next word or idea emerges from your mind, or where to expand on it.

It's all written using the Warp terminal. I don't think it's the right tool for writing a book. I don't know what the right tool would be either, if there even is one, but it's the one I know and the one I have. In the terminal, you can't click with the mouse. You just move around with the arrow keys. Everything is slow, linear, with no shortcuts. Almost like a typewriter, but on a 21st-century black screen.

The ideas for the book come at night. I don't write them down; I try to go back to sleep. But during the day they pop up again. As if my brain had been working while I was resting. You fall asleep without the answer and wake up with it. You don't know where it came from. You didn't look for it. It just appeared, as if someone had left it on the nightstand while you were sleeping.

The key idea that came to me at night was how to structure the editor's folder and file system. That's when I realized it could help me identify the points where I could expand my narrative, where I could delve deeper into the story, where I could describe the experience in greater detail. It was magical to do that.

One of the things I did with Claude was an analysis of the 2026 literature on AI. I wanted to see what others were writing about, what topics they covered, and where the gaps were. I just wanted to make sure I wasn't writing something that the literature already covered. I realized that, in their eagerness to sell, most people end up writing about the same things, as if aiming for a bestseller. And I also realized that you have to write according to your abilities. I'm not a journalist, a college professor, a philosopher, or a writer. But I wanted to tell my story.

My answer is the title of this book.

Keep Pushing. Keep pushing. I don't know if we should stop or speed up. I don't know if we're heading in the right direction or getting lost along the way. But I do know we can't stand still. Standing still in a race is the same as going backward. And this race waits for no one.

Writing the book in a month is imperative, because a month after writing it, half of what I've recounted might already be changing. If I take a year, this book will no longer be useful. So there's no other option: I have to write at the speed of AI. And if something becomes outdated, let it remain as a testament to a moment in time. Because that, too, has value. History books don't cease to be valuable just because the world changed after they were written. On the contrary: they're even more valuable, because they capture a moment that no longer exists.

A new website

About four days ago, on Saturday, while I was finishing watching the Formula 1 race and seeing Franco make a mistake and miss out on points, I thought it was a good day and a good time to build what I planned to be the book's official website. I had been researching on the LLMs (AI models) I use daily about the best way to publish a book. The research led me to a couple of options, but I chose the least common one: self-publishing — meaning not using a publisher for this. At least for now, I didn't want to go down that road; I don't know how they operate, and I think it would slow down getting the book out there.

From the very first moment, the only goal was to share it and reach as many readers as possible, so it could help as many people as possible — that was the goal from the start. So the option of getting a publisher, publishing it, and selling it stopped being appealing. Sitting in front of the TV during what was left of the race, with the sound muted, I opened my Mac and started on the task of having a complete website with all the book's content, so it could be read from any browser or phone. I researched all the websites of all the books I had as references, looking for patterns, but honestly I didn't find anything like what I wanted. Clearly, making an entire book public isn't the basic principle of writers who publish to earn royalties. There is the option of publishing on some sites where they let you set the price to zero, almost free, but you have to go through an official bookstore, register, and do everything the same as if it were worth 10 dollars. With all this information, I confirmed that my path was more toward a website than a publisher and posting it on Amazon via Kindle.

It was around 9:30 AM when I asked AI my first question — I wanted to know which domain and hosting company was the most recommended for starting a basic web project. Among the options they gave me, I landed on Cloudflare, which I actually did know, but not for selling domains and hosting infrastructure. I started researching whether it offered any free hosting solution, and AI confirmed that yes, if I bought a domain I'd have free access to a bunch of useful solutions for publishing the book's website. With that in mind, I signed up on Cloudflare's website and started managing the domain — that means checking if the domain is available (keep-pushing) and with which extension (.com, .ai, .co, .org, .net, etc.). That part was very quick; the domain was free and the .pub extension was very affordable — you pay annually. That's how keep-pushing.pub ended up registered under my name.

With that sorted, I started studying the website's aesthetics. ChatGPT helped the most there, giving me very thorough specifications for how to design a complete website that was technological, clean, and simple for reading a book online. I copied and pasted all that information into the Claude Code terminal so it would generate a new folder with the project and a first file with all the specifications.

Then I asked Claude to put together a step-by-step plan for me to follow, going from zero to having the website ready and published for everyone to access — that was in another file called plan-web. All information provided by AI; I just asked and clearly explained what I wanted. In that plan, it first had me connect the purchased domain with a new web project inside Cloudflare. I followed all the steps and got it done quickly. Then the plan said I needed to connect my GitLab account to the project as well — that saves a lot of clicking in the future. Cloudflare has its own integration for this, so it was just a couple of clicks and it was connected. The next step was creating a GIT repository (where all the code for all websites on the internet is stored). You create it within Claude Code — it knows exactly how to do that, I just asked. Once created, Claude wanted to test that everything was OK, so it created a basic web page that just said "keep pushing" as text in a corner. The goal of the test was for the code of the new website on my computer to reach GitLab, and since that's connected to Cloudflare, it would arrive there too — and when that happens, it means anyone browsing the internet and visiting that website would see the content we'd sent: just two written words. The test worked perfectly.

With that, we already had almost all the pieces needed to create a complete website that anyone on the internet could read and see. Up to this point, I think about an hour had passed, and I remember going to get my second coffee of the morning. Now I was at the part where I had to start building the website I had imagined, so we started on that. Right there in the Claude Code terminal, I told it to review the specifications file and see if it looked good. That took just a few minutes, and then it raised the question of which technologies to use, showing me several options and paths forward. Since there were many, I told it to choose the most optimal one for the project and justify why. I read some of its responses, but without being entirely sure, I followed its recommendation to use Astro — a framework designed for content-heavy websites, perfect for publishing a book.

With that decided by AI, I told it to start creating the first version of the website and show it to me in the computer's browser — Chrome, which is what I use. The great thing about this is that you can review everything, see if you like it or not, and make changes that only you see on your computer. When it finished, I was very pleasantly surprised by what it had done — it was exactly what I wanted. The AI had perfectly interpreted what I wanted to do and how I wanted it to work. Remember that ChatGPT put that part together, so thanks to that, I had version one of the website in about 2 hours. It was completely empty with placeholder content, but its parts worked and I liked how it looked overall.

The next step, before feeding it all the chapters, was to review everything — change, enlarge, or shrink parts or sections, make sure all the links took me to the right place when I clicked. That's how the website started to take shape, and I adjusted how the navigation would work. I remember spending maybe another 2 hours tweaking, changing, reviewing, and tweaking again. Most of the time I'd send a screenshot and say "I don't like this, make it bigger and align it with this text or this screen" — Claude Code is perfect for that, it does it very well, it doesn't make mistakes interpreting those instructions. When the website was well advanced structurally and everything was in its place, I started feeding it the content of 2 complete chapters, 1 and 2, to see how everything looked. For that part, I had to use another tool: the translator at www.deepl.com, a company with an online translation service. I had to subscribe to a plan, but they gave me 30 free days, so I started feeding it all the texts to use the translations in the 2 remaining languages I'd chosen: English and Portuguese. I chose English because I'd studied that when models are trained, 90% of the text they use is in English, so I figured it might be easier for the models that are constantly using their bots to read new content to pick up my book and use it in some future training. I added Portuguese because I've been working with a company based in Brazil for 3 years now, so Portuguese would help me share the book with them too.

The translation and chapter generation process for the website must have taken another hour until everything was organized. In this part, we had to regenerate the code we had for the Spanish version for the other 2 languages and verify that everything worked the same as the Spanish version. I had to do a lot of clicking and testing, but everything ended up working perfectly. With that, we had 2 chapters in 3 languages, all loaded. It was a big step forward, and it was probably close to noon. I don't remember having lunch, but I do remember being alone at home on the couch, drinking coffee and eating some snack. When I reached this point, the plan said it was time to create the contact form, so we did that part. AI asked me if I was going to use a mail server so I'd receive the emails, but I'm not a big fan of that model, so I asked AI for other options. After a couple of questions, it suggested having a database where all form submissions would be stored — which seemed perfect to me. Whenever I wanted, I could check who filled out the form and reply to them.

To achieve this, it suggested using Cloudflare's free plan, which already has a database (DB) ready to use on my website. It configured everything and took care of running the tests to make sure everything worked. Since we're connected to Cloudflare via API, I didn't have to click anything — it did everything through the API. I just gave it a token so it could connect to my account and configure everything. With the contact form done, we moved on to the one that asks for your email to receive book updates — sometimes people fill those out on sites where they want to stay connected with news. After that, I had an idea: it would be great to be able to select any part of the chapter text and leave a comment. So I asked if something like that already existed, and it said no, but it could build it and I should see how it turned out.

A few minutes later, it asked me to try it on my computer. So I went and selected a piece of a paragraph, and as soon as I released the mouse with the text selected, a little window popped up asking for an email and a comment. Fantastic, I thought — this way anyone reading the book who has a suggestion or something to contribute to the discussion can do it right on the piece of text they want, and the complete information would arrive in my database: the 3 parts — the paragraph text, the email, and the comment. With that resolved, the structure was almost done.

Another important step was writing the book summary text, my bio, and other texts I didn't have yet. I spent a couple more hours on that, and it was getting close to 7 PM when I realized — because someone arrived at my house and asked: "You're still at it?" So in less than 10 hours I had everything ready, published on the internet, and ready to be read — at least those 2 chapters. The next day I had a bit more time to review everything and fix some last-minute errors. I was able to make a logo and a cover design using Canva's AI application and upload a profile photo. With all that, I mustered the courage to take the first public step, so I sent a message to my friends — the same ones from the fishing trip — telling them about the new book. My message was met with a flood of congratulations that felt really good and justified all the work done up to that point.

A couple of days later, I had time at home in the evening to process all the corrections and notes my editor had left me, translate all the remaining chapters, and publish everything new on the website. It got pretty late that night, but it was worth it because by then the book was almost finished and published. From when I started to that point, about 15 days had passed, which was an optimal timeframe for the plan of having it ready in a month. I have 15 days left to keep polishing and completing some ideas I still have.

One of the things I also did was let my cousin know I'd written the book, so he could read it and help me make corrections, and also ask him to write the prologue — to which he replied that he was happy to do it and help me. It was great news because if there's something conceptually poorly explained, he'd catch it.

The process of writing keeps appealing to me more and more. Now that I've finished, I'm already thinking about whether it would be possible to write something else — maybe about Artificial Intelligence, or maybe more about strategic business management, or telling some other story. For now, I've only begun to slowly let my family, friends, and acquaintances know about the book's existence, to get some feedback that might help me revisit some ideas or passages.