The Bubble

Inspiration strikes again. Now surrounded by greenery and nature, with the perfect view from the backyard patio. The house is empty. A cup of freshly ground coffee that always gets my mornings off to a good start. Spring is here; there are two days left until the 21st.

Everyone is silent, except for my mind. Pure inspiration. The most productive day: twelve hours and counting. Electronic music is key, too.

Looking up at the sky, watching the birds feeding in my garden, my neighbor's pets, my cat. And in that moment, something clicked.

The animal kingdom is oblivious to the latest developments in AI. Birds aren't going to stop eating seeds just because a language model just broke a benchmark record. My cat doesn't know what a token is. And millions of people, given the lives they lead and their jobs, will never come into contact with an AI terminal.

That's what sparked the idea.

Humanity's evolution has been shaped by kings, rulers, artists, geniuses, generals, warriors, institutions, companies, and corporations, by key figures who transformed daily life on this planet. But the vast majority of people watched historical events unfold without being able to intervene, change them, or even understand them. They were simply spectators. That's how it was with the printing press, with electricity, and with the internet. Those who invented, those who adopted it early, and those who one day woke up to find that the world had already changed.

The vast majority of humanity didn't wake up today thinking about artificial intelligence. They didn't wake up worried that their jobs might be at risk. They don't closely follow what world leaders in artificial intelligence are doing. For most people, today was just another day.

AI is a concern for only a few people. Perhaps it's like writing in its early days, a revolutionary invention that for centuries was used only by scribes and priests. Perhaps it's like science, once reserved for those with access to laboratories and universities. Like the conquest of space, which even today is the domain of only a few countries. Or like the discovery of the Americas, where an entire continent existed without the other knowing it.

Great transformations always begin in a bubble. A small group that sees something others don't. And then, slowly or suddenly, the bubble expands and the world changes. The question is how long it takes for the bubble to stop being a bubble. And whether, when it expands or bursts, it will be for the better.

It's very difficult to predict when real change will come. When will that 93% of the world's population that hasn't adopted AI in their daily tasks begin to do so? Writing has spread to 88% of the world today, and we've had a little over 200 years of modern writing history. We still haven't reached 100%.

When will it go from 3% to 5%? Can it be predicted? How fast is it actually advancing? No one has the answer. And anyone who says they do is just guessing.

Let's start with the basics: if you don't have a laptop, internet access, and time, you won't be able to get started. It's very difficult to grasp this reality using just a cell phone. If you don't have a school or university to educate you, the road ahead is even longer. If you don't have a supportive family environment, it's even more challenging.

AI isn't just a matter of technology. It's a matter of access, education, and context. And in all three of these areas, the world is profoundly unequal. A 15-year-old in Mendoza with a laptop and Wi-Fi has more access to AI than a 40-year-old professional in a rural town without a stable internet connection. The gap isn't just technological. It's geographic, economic, and generational.

I'm inside the bubble. Among the top 3%. By my own choice and because my world has always been software. Ever since I was 10, when I got my first computer, a TK-85, the Brazilian clone of the Sinclair that dominated Latin America. My dad gave it to me, and we shared it, too. He was a fanatic just like me. Maybe even more so.

The TK-85 was a black box with a membrane keyboard that you connected to a black-and-white TV. You programmed it in BASIC, saved it to cassettes, yes, audio cassettes, the same ones you used to listen to music, and when you loaded a program, you had to wait minutes listening to an unbearable noise until it finished loading. Sometimes it failed, and you had to start over.

And yet it was magical. I programmed contact lists and solved other basic problems.

IF, IF ELSE, LET= I still remember them. Every line of code was numbered. Line 10, line 20, line 30. If you made a mistake, you had to rewrite the line. There was no mouse. No menus. No undo. It was pure logic and patience. And from that moment on, something clicked inside me that never went away. By the time I was 18, I was already creating databases in Visual Basic to store my college notes. They had no real practical use, but I was always drawn to the tools themselves.

Some people collect cars; others collect wine. I've always collected ways to organize information.

In April 2026, active users of generative AI were estimated to be equivalent to about 29% of the world's population. The remainder was around 71%: nearly six billion people. The technology seems to be everywhere, but its use is still far from universal.¹

But there's another group that I find even more interesting: those who've already tried it but haven't yet incorporated it into their daily lives.

That, to me, is where the next big wave of adoption lies.

The Gap

These are people who have already crossed the first hurdle. They know ChatGPT exists; they've asked a question or two, maybe generated an image, or solved a specific problem. But they haven't yet discovered what it means to have artificial intelligence available every day for working, learning, researching, writing, thinking, or simply saving time.

When the topic comes up in conversation, I try to show what it can do. I often offer my help, too. But a single conversation is rarely enough.

There are people who are just one conversation, one video, one demo, or one book away from really starting to use this technology.

This book is intended, in large part, for them. And if you're reading this, you've probably already taken that first step.

ChatGPT reached over 100 million monthly active users just two months after its launch. Until then, no consumer app had reached such scale so quickly, faster than Facebook, Instagram, or TikTok. But the number isn't the most important thing.

Looking at that figure in hindsight helps us understand where we stand. We can talk about 2030 as the year when nothing will be the same as it is now. In the first few months of GPT's existence, Google declared a red alert. They relied on traffic and the searches everyone was making, and out of nowhere, those numbers began to drop. People preferred to stop Googling and start "GPT-ing."

Sundar Pichai summoned Larry Page and Sergey Brin—the retired founders of Google—to get back in the game. Years later, Gemini was released. But the damage had already been done.

Dozens of models were released. Each version represented a massive leap in quality over the previous one. Unlimited investment poured in. They created open source models, and from there, anyone could build upon them. Recruiting talent became critical: everyone was changing jobs; very few people truly understood how AI worked, which is why they were so sought after. Comparative benchmarks of each model's intelligence made headlines every day.

In one year, what normally takes ten years happened. Never in the history of technology has there been such an intense, fast-paced race with so many players competing at the same time.

AI companies were copying one another. And we watched it all from the sidelines, like spectators at a boxing match where every round brought a new champion. We were the ultimate beneficiaries, as we had access to a superior model with each passing day. With every month that went by, what you'd learned the previous month was already outdated. A prompt that worked perfectly for you in March stopped working in April because the model had changed. A technique you'd mastered became obsolete because a better one came along. It was exhausting and fascinating at the same time.

The Race for AI

In this battle, all the AI companies ended up fighting each other. And there's a lot at stake. The one person all the companies had to befriend was Jensen Huang—the man in the leather jacket, CEO of Nvidia. Out of nowhere, he became the most valuable asset on the planet, because his company manufactures the GPUs—the chips that perform thousands of mathematical calculations in parallel—exactly what neural networks need. All the AI you use runs on Nvidia chips. No one sees them, but without them, there's nothing. A component that most people don't even know exists is what makes everything we use every day possible.

And it's not just the hardware. AI software is built on CUDA, Nvidia's platform. PyTorch, TensorFlow, all optimized for Nvidia. If that changes, everything has to be rewritten. Nvidia is the master of AI today. They control both the chip and the software that uses it. It's as if a company owned both the roads and the cars.

But let's keep playing TEG a little longer, because this is a game of global geopolitics and every piece is connected to the others.

Taiwan, an island in the Pacific with 23 million inhabitants, manufactures 90% of the world's most advanced chips. TSMC (Taiwan Semiconductor Manufacturing Company). Imagine being the president of Taiwan for a few weeks. The whole world needs you, and the whole world is watching you. Without TSMC, there are no iPhones, no Nvidia GPUs, no AI, nothing. And the machines that manufacture those chips are only made in the Netherlands, by a company called ASML.

The future of technology depends on two places that most people couldn't locate on a map. The major powers have clearly set out on the path to self-sufficiency, though they're unlikely to achieve it quickly, there's a 20-year experience gap between them. The United States has Nvidia building factories in Arizona. China has started building its own. To wrap up this TEG board, we've added to the map the countries that produce silicon, because silicon is to the digital age what iron was to the Industrial Revolution. And the world's leading producer of silicon is China.

Speaking of China. An AI lab in Hangzhou, founded in 2023 by Liang Wenfeng, a billionaire with no background in AI, launched DeepSeek. These models are highly competitive at one-fifth the cost of the competition. It's not perfect, but it works just fine. I use it, and after tweaking the prompts, it performs well. It still struggles with tool selection within its decision-making flows, but the latest version showed a remarkable improvement. This really sped things up, because it proved that you can compete even with less investment. Silicon Valley panicked.

Nvidia's stock fell. Investors began to question whether billions of dollars were really necessary to train a good model. If a Chinese lab could achieve something comparable with a fraction of the budget, what were the others doing wrong? Were they overspending? Or was DeepSeek hiding something?

As an aside, I'll tell you that I received an email from OpenAI that said this: "We're reaching out to you to better understand why some teams are reducing their usage, suspending their activity, or changing the way they use the OpenAI API, and what factors most influence those decisions." Now they're worried, because my bill went down. I didn't get a single email when one of my keys was hacked and I was charged about $500 because someone used it on image-generation models. That's just how it works.

On February 2, 2026, Elon Musk merged SpaceX with xAI. Global connectivity with Starlink. The ability to launch data centers into space so that maintenance and cooling costs drop to nearly zero. Grok, his own AI model. And X (Twitter) as a source of real-time data. Rockets, satellites, artificial intelligence, global data, all under one roof.

No one has ever had simultaneous control over so many layers of critical technology. Rockets to reach space. Satellites to cover the planet. Chips to process data. AI models to think. And a social network with hundreds of millions of users to feed all of that with real-time data. It's the greatest concentration of technological power in history. And there are no signs that this will change anytime soon.

SpaceX bought Cursor for $60 billion. Nvidia is in talks to buy Hugging Face for $13 billion. The figures involved in deals for strategic AI assets are mind-boggling. Every acquisition redefines the power map.

In 2026, Jeff Dean left Google to found Discovery Loop alongside Sanjay Ghemawat, Oriol Vinyals, and Quoc Le. The startup aims to use AI to accelerate research: propose hypotheses, run tests, analyze the results, and use what it learns to design the next experiment. They are starting with machine learning research, with the intention of extending that approach to other scientific fields. When people with that track record leave a company to do something new, the world should take notice.²

Latin America has always lagged behind in technological developments. Brazil has always been at the forefront, but when it comes to AI, it hasn't yet established leadership in the region. The race is always told from the perspective of Silicon Valley, China, and Taiwan. We, on this side of the world, watch it unfold and try not to fall too far behind.

How do I follow this race from Mendoza? I subscribe to three different newsletters, and on X I'm connected to experts' accounts whose content I review. I don't have as much time as I'd like to keep up, and besides, it's sometimes counterproductive because things are constantly changing. They come and go and come back. Not everything that seems like it's going to dazzle ends up shining, and perhaps what didn't make it anywhere is what's truly important. Ultimately, you have to find a balance here.

Software belongs to everyone

For the first time, what had been predicted to happen actually happened.

One of my partners asked me, "Hey, did you see this?" and sent me a link. I'd never seen it before, so I clicked through to see what it was about. It was the GestionaISP website, software for internet service providers, our very own market. And the URL had a "/vs/wispro/" in it. Versus Wispro. Versus us. I spent maybe 30 minutes looking at it.

The URL wasn't a coincidence. AI models thrive on comparisons: what's better than what. That URL was designed so that when someone asks the AI, "What alternatives are there to Wispro?", GestionaISP would appear first. Well thought out. Very well thought out.

A new phenomenon was born: AI-friendly web design. Now you don't just have to convince Google, you have to convince AI models to recommend you.

The entire website was built to be read by AI models. Lots of explanatory text, clearly highlighted concepts, clear URLs, and structured data that tells the models exactly what kind of product it is, who makes it, and what frequently asked questions it has. 90% of the content used to train the models is in English, and only 2% is in Spanish. It's easier to translate later than to train the models twice.

They had free access to a demo. It looks like it was coded or programmed entirely with AI, what's now called "vibe coding" or "prompt-based coding," where you gradually tell the AI how you want your website to be.

VisualNet was a Wispro client for years. They created Gestiona ISP. We started with 30 clients and grew to over 700 before they stopped being Wispro clients. On their website, they explicitly state that Wispro had been very helpful to them and that they had no complaints.

They know us inside and out. And I'm proud that my company serves as a foundation for the creation of new ones. Does it worry me? Yes. Does it make me angry? No. It's exactly what I've been saying throughout this book: AI gives everyone the tools to do what previously only a large team could do. And this time, those tools were turned against us.

There's a lot of talk about the SaaS (Software as a Service) apocalypse, also known as subscription-based software. The phrase is shocking, but it's not an exaggeration. The product is no longer the differentiator. AI generates better code than humans in most cases. With good specifications and a team of specialized agents, you'll be able to pull it off. That's vibe-coding. And it's no longer just theory.

Many traditional software companies must be feeling threatened. It's very likely that a single person working at a company, someone who knows a process inside and out and has contact with customers, could start their own company and become your competitor overnight.

No one is safe. And it's not that person is disloyal. It's just that the barrier to entry no longer exists. Before, you needed capital, a development team, months of work, and thousands of dollars. Now you need an idea, a prompt, and time. Loyalty competes with opportunity, and opportunity has never been so accessible.

Software is no longer the differentiator. Service and support are what are beginning to set companies apart. People are talking about the era of "Service as a Software": embedding yourself within the customer's operations and taking charge of the entire execution. It's a huge challenge. But the path is there. And anyone who fails to see it will end up like those who didn't notice GestionaISP until a partner sent them the link.

We're no longer going to be able to sell the same software to 2,000 customers. Even if they're in the same industry, their organizations are very different: size, corporate culture, processes, revenue models. There are too many nuances for everyone to want to stick with a single solution. And yet, that's exactly what we did for 20 years. One product, thousands of customers, and each one adapting as best they could.

If the product is now easy to develop and there's nothing stopping others from copying it, why not reach each customer with a version customized for their use case and business model?

Last night, at a meeting with colleagues from tech companies in Mendoza, we were discussing exactly this. There were about fourteen of us, but five or six were deeply into AI. One of them said he was already working on preparing for these new challenges.

A colleague shared an article titled "SaaS Isn't Dead. What's Dead Is Uniformity." And it explains precisely the shift that software development is undergoing today, where companies no longer want a generic solution — they want something customized for their processes and their business. AI is exactly what makes that possible: the ability to generate software at a reasonable price, moldable to each company that needs it.

There's also talk that charging per screen and per seat, where the user pays just to sit there and click, won't work anymore. Because an agent will connect to the data and APIs and perform the same tasks a human used to do. The software will be used by AI, not by people.

If your customer no longer needs the screen, why should they pay for it?

The Evolution of Software

In early August, I had an epiphany. I started thinking about a new way to develop AI-first software. The idea kept circling in my mind for several days until I was able to put it into writing. It often happens to me that I have things swirling around in my head; then I sit back calmly to think about what it might be, and that's when I find the starting point and begin to connect the dots and reorganize everything to give shape to the idea. Only when I have a clearer picture do I try to put it in writing. It's a process I've always followed: mulling over ideas.

The concept is radical yet simple: a single software instance per client, with its own backend, frontend, and integrated AI that allows you to generate software from within. The client doesn't wait for the development team to build what they need. They build it themselves, by conversing with the AI. If they want a new view, they ask for it. If they want a dashboard with metrics specific to their business, they build it. If they want a report that cross-references data in a way that never existed in the original software, they generate it. Surely some companies have already started down this path to offer these kinds of solutions.

The more you customize it, the harder it would be to walk away later. It's the smartest defense against competition, not building walls, but roots.

I created an evolutionary pricing system for it. It starts with a minimal entry fee. As the software grows, you consume and pay for the tokens required by your customizations. We add a percentage of that consumption to the monthly price. If a customer was paying $10 and customized their version by spending $120 on tokens, we add 10% to the price: it goes from $10 to $22.

The more you customize, the more you pay. It makes sense. And someone who's been developing software for a specific industry for 20 years knows it inside and out. They're more likely to understand their customers and their requests than any new competitor.

I did a lot of research to make sure I didn't repeat any names. I spent about two hours brainstorming and tried many options until I said, "This is it." When I saw it, it was instant.

In the end, I called it Vibe-Shaping. Because it's not vibe-coding, you're not building software from scratch. You're shaping something that already exists, molding it for each client. Shaping. Giving it form.

We talked about the business, the competitors, the market, new models, and opportunities. But there's something we haven't touched on yet, and it might be the most important thing of all: what's happening to us? To the people who are in the midst of this revolution. Because you can have the best business in the world, the best AI, the best Vault. But if you don't stop for a second to think about what all of this is doing to you on the inside, you'll miss the most important part of the story.

1. Kemp, S. (2026). Digital 2026 Mid-Year Global Update Report. DataReportal. Estimates 2.42 billion active users of generative AI in April 2026, equivalent to 29.2% of the world's population. It cautions that the figure may not represent unique individuals. The remaining percentage does not establish how many people have never tried it. https://datareportal.com/reports/digital-2026-mid-year-global-update-report ↩
2. Radical Ventures. (2026). Discovery Loop. Overview of the company and its founders: Jeff Dean, Sanjay Ghemawat, Oriol Vinyals, and Quoc Le. Describes its goal of automating research cycles using AI, starting with machine learning. https://radical.vc/portfolio/discovery-loop/ ↩