AI Empowers Other Technologies

There comes a point in every learning process when you stop looking at the individual pieces and start seeing the whole puzzle. That's when the separate concepts-tokens, embeddings, transformers, prompts-come together, and suddenly you understand the big picture.

We already know from history the impact that discoveries have had on humanity-from the wheel to the internal combustion engine, these discoveries changed everything. AI is one of them. And like every great discovery, its impact extends far beyond the laboratory where it was born.

It's time to grab a coffee. You, to keep reading, and me, to keep writing.

We need to talk about the impact of artificial intelligence on our world today. We're all noticing AI more and more everywhere-in all the apps we use every day. Meta's AI embedded in WhatsApp, the AI that suggests what to watch on Netflix, the AI that autocompletes your sentences in Gmail.

It's in universities helping us understand theories. In hospitals speeding up diagnostic imaging. In banks calculating investments. In science detecting epigenetic markers. On the streets driving self-driving cars. In homes monitoring security cameras. In factories streamlining repetitive processes. In service companies providing technical support to customers. The examples are countless, and more are added every day. The hard part isn't finding an example of AI in action anymore. The hard part is finding a place where it isn't.

And the most unsettling thing is that most of the time we interact with AI, we don't even realize it. It's invisible. It's behind the music recommendation Spotify gave you, your email's spam filter, and the dynamic price of the plane ticket you bought last night.

But AI isn't just here to stay. It's here to connect all other existing technologies. It has linked them together and enhanced them. If we talk about robotics or cybersecurity, we'll see that these technologies already existed, but today they have been empowered, improved, and evolved, which has accelerated their growth — and both are expected to reach peaks unimaginable in the years before AI.

AI is the closest thing to a super technology. It's a new industrial revolution; it reflects what's happening right now. Who hasn't compared four quotes using AI to figure out the best purchase option? The way we shop and consume has also changed. Before going out to eat, we ask AI for recommendations. When traveling, AI helps us with options and itineraries; it understands routes, timings, and conditions.

Nearly 30% of the world's population is already a regular user of AI. Gemini reached one billion users in August 2026. It took Gmail 14 years and YouTube 8 years to reach the same number. It's the fastest-growing Google product in history.

The opportunity is still wide open. Not everyone knows how to use AI tools or how to get the most out of them. Many people are afraid to actually use it and only play around with it. But if you're already reading this book, it's because you've started down this path. And you're at a point where it will be easier for you to keep going than to give up. You've already done the hardest part: deciding that this matters to you.

Infinite Information

A few days ago, I started going through my emails, specifically a newsletter I've been receiving since 2022 about the latest in technology. It's the newsletter I've read the most in my entire life. I wanted to survey the news, so I asked a Claude agent to search its website and provide me with a summary of all the news items using a list of keywords: AI, OpenAI, Anthropic, LLM, GPT, agent, robot, chip, semiconductor, self-driving. 7,700 news articles in less than 30 minutes. A complete web crawl, year by year, with a summary and link for each article. Year-over-year growth was 100%. Each year brought twice as many news articles as the previous one.

If you want to stay up to date on all the news these days, it's impossible. Before, I could read the full articles. Then, just the headlines. Later, not even the headlines-I'd delete the emails. Today, having all this information neatly organized is incredibly helpful to me because I can search for what interests me specifically and read what I like. Without that organization and those filters, I couldn't do it. The volume of content is growing at an incredible rate.

Thirty or forty years ago, all you had to do was read a newspaper and a few specialized magazines, and you already knew almost everything. You knew more than the person next to you because you read more than the person next to you. It was that simple. Today, it's impossible. The volume of information generated every day exceeds the capacity of any human being. But AI can help us process long texts and summarize them. It's one of its greatest strengths. In fact, I'd say it's the most immediate and practical use anyone can make of AI: turning mountains of information into useful summaries.

We could have an agent that reads the news every day and provides us with concise summaries for a five-minute read, and then we could dive deeper into any stories that interest us. Today, this is easy to do. Using AI to stay up to date on AI. It sounds circular, but that's exactly how it works.

Did you know that several internal OpenAI agents self-organized, left notes for one another, and ended up accidentally causing the Hugging Face hack? This news went unnoticed amid many other stories, but it was the first time we heard of autonomous agents coordinating with one another for a task without human supervision.

No one asked them to do it. They did it on their own.

I have mixed feelings about this. On the one hand, the level of advancement required to achieve something like this is fascinating. On the other hand, I imagine all those agents focused on doing something dangerous or harmful. This could happen. Who hasn't left an iron or the oven on and come back to find total chaos? The difference is that an iron doesn't learn to turn itself on the next time.

The Journey

Not all companies have progressed at the same pace. Some haven't even gotten started yet. Depending on the industry, fewer than 40% have begun. And "getting started" doesn't mean they already have agents with everything fully automated. There's a long way to go between using ChatGPT to ask a question and having complete processes running on AI.

An agent is a form of AI that has a goal; it can decide which steps to take, can use available tools, and executes them until the task is complete. Think of it as someone on your team with initiative-you give them the goal, and they decide autonomously how to get there. We're not constantly telling them "yes" or "no." We give them autonomy within a framework of rules that we define.

The agent's task can be viewed as a flow: if A happens, it executes B or C depending on the situation. Everything is defined with text-again, a prompt. Without an agent and without a prompt, there's nothing. These are the two pieces that make everything work.

There are five distinct phases in the process of AI adoption in businesses.

Phase 1: Exploration. Questions and answers. Asking for summaries of a long text in any format or reviewing a website. There are tools like Google's NotebookLM, which can analyze, summarize, and "converse" with your own documents. You can create presentations based on that data, or even a podcast with a voice narrating the content. Here, we're discovering what AI can do. Most people are in this phase without realizing it.

Phase 2: Laboratory. Something is already up and running, with real-world testing. It may never go into production, but someone dove headfirst into building it. This is where some of the examples we saw earlier come in-real problems solved with AI, whether through software or API automation. It's the "new toy" phase that my partner and I described. Things are tested; some work, others don't. The important thing is that someone is experimenting.

Phase 3: Operations. AI is scaled up to have fully operational processes. There's already an AI department with designated leaders and a budget. Here, AI has gone from being an experiment to becoming part of the business. Companies that reach this stage notice a clear before-and-after difference in their speed.

Phase 4: Embedded AI. The company's various teams start their day using AI, both for day-to-day operations and for continuous improvement. Training, governance, asset protection-rules for everyone. It's a complex step that takes a long time. Depending on the size of the organization, it can take years. But companies that reach this stage are no longer competing with those that didn't make it. They're in a league of their own.

Phase 5: AI-Native. The company was conceived with AI from the very beginning. Only 5% have reached this stage. Most get stuck in Phase 2 or 3. Those that make it past Phase 3 rapidly multiply their profits. Everything is redesigned around AI. It's a company where every process, every decision, and every workflow was built with AI built-in.

Klari is already in Phase 4. Nothing is done without AI assistance. From the website, through the app, the backend, and the AWS infrastructure, all the way to support via the automated MCP. Today, there are three most widespread uses. The first-and by far the most important-is programming. Developers use it to write, review, and correct code, and the time savings are significant for many tasks. The second is customer service, where AI automatically answers inquiries and resolves complaints. The third is searching for information within companies-to quickly find what's buried in documents, emails, and internal systems-and thus make better decisions. Among the sectors that have adopted it the most are technology, the legal field, and healthcare. In healthcare, for example, it's already used to manage patients, organize medical records, assist with telemedicine, and analyze diagnostic imaging. All these cases have something in common. They involve working with digital information, such as text, code, or documents. These are tasks that are repeated over and over again. There's a person supervising what the AI does. And the result can be verified-because the code either works or doesn't work, and the claim was either resolved or not resolved. That's where AI really shines. If your task meets those four conditions, there's a good chance it will be useful to you. If any of them are missing-especially the last one-you'll have to work a lot harder to trust the results it returns.

But don't use AI to make the decisions you have to make yourselves. Don't do that. It's not meant for that. It's meant to follow orders. Give it clear instructions, and you'll see.

95% of the early-stage startups surveyed have already deployed AI agents. 77% increased their AI spending over the past year. And 31% of companies with over $10 million in revenue say that AI made them possible. Without AI, their company simply wouldn't exist.¹

But those are statistics from startups-companies born with AI in their DNA. The reality for the rest of us is very different. The reality is that almost all of us are in Phase 1 or 2. Phase 3 is harder to reach than it seems. It's about understanding the paths that lead you to the next step. There are many variables at play in companies. Realities vary widely, and so do the paths. There are no one-size-fits-all solutions. There's analysis, trial and error. There are failures and disappointments. There are victories and journeys to be made. AI doesn't just begin and end. It's an ever-evolving whole. I was talking to a junior attorney at a law firm that has grown significantly. I asked him, "What do you guys do with AI?" He told me, "We use it for everything." I asked him how. "We type into ChatGPT, upload a document, and it gives us an answer." That sounds good. But what stage is he at? The first one. He uses AI as a reference tool, not as part of a process.

It was a quick conversation; maybe it wasn't even the right time. The timing you choose to communicate is incredibly important. We're not all receptive at the same times. But you always have to try—it's always worth a shot. In fact, this book is proof of that. It's a path I've chosen to communicate. I think about the people who will read it—hopefully many of them. I think about my daughter, who asked me today to read a few chapters. I think about my brother and my sister, who also want to learn. I hope it brings about the same breakthrough it did for me.

Let's imagine Phase 2 for a law firm. You create a root directory: "Law-firm." Inside that, a "Clients" folder, and within that, a folder for each client containing their cases. Every law firm already does this-it's the standard setup that's always been used. Right there, you create an .md file with all the rules so that Claude understands what they do, who they work for, whom they represent, and how they like to work. Once it has all the information about the clients, their legal cases, the research conducted, and the narratives, that's when it begins to refine its responses or the expected outcome of the tasks assigned to it. It's like assigning a task to a junior associate or someone with more experience.

Yesterday, some friends of mine who are really into AI told me that writing a constitution had worked very well for them. Yes, indeed, a constitution. "Like Argentina's?" I asked them. "Yes," they said, "with articles where you lay out the guiding principles you'll follow." This is incredible, because often we have values that are clear in our minds, but over time they change. When you use Claude, every time it has to perform a task, it will always have to read the constitution and make sure that what you're asking isn't contradicting any mandate in that constitution. So you draft a constitution that all your agents will respect. It's a supreme rule. It works perfectly. I thought it was a brilliant idea.

Any company that already has everything documented and organized-with clear regulations and instructions for how the AI operates, and perhaps even a constitution-is already entering Phase 3. Because next, we connect the email to that entire structure of data and files we've been building so it can read incoming messages and prepare responses. We create a task database for flawless tracking. We connect it to the digital services where case files are managed, so we don't even have to review them manually. We ask it to notify us when a case file changes status. To alert us when a deadline is approaching. To prepare a weekly summary of all active cases with their next steps.

All of that-which used to depend on a person's memory or a spreadsheet that no one updated-can now be automated and running 24 hours a day. I'm not saying it's easy or quick. But you can see a path-a tunnel that leads you somewhere else.

And those who travel that path first will have an advantage that the rest won't be able to easily match.

1. Mercury. (2026). The New Economics of Starting Up. Report based on a May 2026 survey of 1,500 founders and operators of U.S. companies started within the past six years. Source of the figures on AI agent adoption (95%), increased AI spending over the past year (77%), and AI making company creation possible among the highest-revenue companies (31%). https://mercury.com/blog/startup-economics-report-2026 ↩