What are you investing your time and money in today that may simply become unnecessary three to five years from now? Websites, content management systems, CRM systems, chatbots, forms, analytics, and dozens—actually hundreds—of different services and subscriptions. The main question now is whether we will need them at all. In this episode I want to show you a real case involving GPT-5.6 Sol and Claude Fable. I have a large active website with an enormous number of pages. I gave these systems the task of rebuilding it, moving it into a new environment, and creating a proprietary content management system around it.
That example gives us a way to examine whether we still need dozens of familiar applications and subscriptions, which technologies will remain relevant in the near future, and how much money and time people may now be wasting—not only on products that already exist, but also on the countless solutions that everyone around us claims to be building. This episode is not about website development. It is about where it makes sense to invest your time, money, and attention. Hello, everyone.
You are watching ToTheMoon, and I am Alexander Volchek. Today I decided to record an episode that continues our discussion about artificial intelligence while also showing you some of the genuinely new things happening in AI and the fundamentally new capabilities that are appearing. I have repeatedly argued that a huge number of cloud products, software applications, and established processes will either become unnecessary or be transformed very deeply. Transformed is the important word here.
I would not build the argument around the idea that everything will be taken away from everyone, all jobs will disappear, and nobody will know what people are supposed to do. Instead, let us look directly at what is happening to specific services and programs. I want to use an example that may not seem spectacular to everyone, but it is a good demonstration of where artificial intelligence is heading and what can now be done with modern models and systems. I will focus mainly on ChatGPT 5.6 Sol and Claude Fable, although much of what I describe was already possible with earlier OpenAI, ChatGPT, Anthropic, and Claude models—GPT-5.5 and Opus 4.8.
One point is extremely important: the capabilities of these models differ sharply depending on your subscription and the access it provides. I have said this many times and will continue to repeat it. To experience many modern models fully, you need access to particular modes. Paying two hundred dollars does not merely give you more tokens. The point is access to specific operating modes. GPT-5.6 Sol, for example, introduced an Ultra mode. Fable has an Ultra Code mode and a Max mode.
Now let us turn to systems called CMS platforms. These are content management systems, the foundations on which websites are built. I first encountered a CMS in the early 2000s, when I was still at university.
At that time, building a website was more difficult in one sense and perhaps easier in another, because the number of available tools was small. Over the following twenty-five years I watched these systems undergo major changes. Naturally, during that period I met many owners of companies that developed CMS products, including people who made good money from them. Today we can see a market full of new systems for building websites. Why am I using websites and CMS platforms as the example?
Because there are hundreds of millions, perhaps billions, of websites in the world—someone can correct the exact number in the comments. This is an enormous mass market and remains a primary entry point into the internet. There are also many mobile applications, of course, but websites make the change especially easy to see. If websites begin to be built differently, mobile applications will also be built differently. You know my view that a great deal of software will no longer be necessary.
Even complex systems such as sales CRMs, and large parts of ERP platforms, begin to look unnecessary in the approaching future as serious tools appear inside Anthropic and ChatGPT. At the very least, their necessity becomes relative. I had a task that you have heard me discuss before: when will it become possible to take an entire website and move it with the snap of a finger, integrate everything easily, make every function work, receive leads, create bots, and so on? That used to look impossible.
Perhaps a year ago, or even six months ago, Ilnar said in one of our podcasts that you could move a page, but you could not yet move an entire website. What has changed? We are now watching the development of models in the Sol and Fable class. What makes Sol interesting over the last few days is that it has been working for me without stopping for three days. I have never seen that before. I saw something close to it with Fable, but Fable no longer seems to work that way; I suspect Anthropic introduced certain restrictions.
It also periodically switched to simpler models because of Anthropic's limits. In this case, Codex running Sol only suggested moving to another model for greater speed. It occasionally said that faster models were available. The large question is whether a model that works for three days, a week, or two weeks is doing high-quality work or is simply slow. We will find out. It is still too early for me to draw a conclusion, because the long-running program has now been working not for three days but through Friday, Saturday, Sunday, and Monday—four days, at least four and a half by the time I record this on Tuesday.
We will see the final result. By yesterday evening it still had not finished; it continued through the night. I kept asking for status updates, and it told me that everything was fine and development was continuing. Several times I tried to expand the work by telling it to launch parallel agents and broaden the process.
At one point it replied: “I am operating at the limit of my capabilities. A large number of parallel threads are already running, and I cannot create any more.” That was its answer. Over these days I gave the models the following tasks. The work originally grew out of a design project after Claude Design appeared. Watch the episode we released a few days ago, where I discussed ChatGPT Voice and Claude Design. I began developing a design in one of my projects inside Claude Design; the process actually started with a real-estate project.
I then moved to a broader project devoted to organizing and developing the large amount of content I own, storing that content, managing my various programs and educational products, my channels, businesses, and partnerships. There are many different elements. It is a large portal, and at the moment it is highly fragmented. Many parts are not connected: Telegram exists separately, WhatsApp separately, several different payment systems are involved, and so on. It is a heavily loaded and diverse environment created and maintained by different programmers and other people.
The fact that it was built by different people is important. I updated the real-estate design in Claude Design and then decided to start my own project in parallel and see what would happen. Here is the central problem. The introduction was necessary to explain it. If I want to generate pages automatically for advertising campaigns—so that I can create thousands of campaigns in different areas that interest me—I need the ability to generate landing pages, full websites, or intermediate sites.
I already understood that artificial intelligence has completely solved the task of creating bots automatically. There is no longer any need to buy a separate system for interacting with bots or messengers or for building communication sequences. This applies to Telegram, WhatsApp, SMS, Facebook Messenger, and essentially everything else. If a system exists, it almost certainly has an API. In the past, multichannel or omnichannel platforms were fashionable: a CRM or another system would connect Facebook, Telegram, text messages, email, and other channels in one place.
The owner of the Moneycat project lives nearby; it is a substantial company, and he and I discussed this many times. I remember those conversations. The company was at one level then and later grew considerably. But now the question becomes: why should you use that company’s product? After meeting its owner, Mika, I installed the system in one of my projects, and it remained there in some form. The commercial teams in my other projects did not want to deploy it. That is not the central point; the product itself was reasonably good and included features I liked.
Today, however, you begin to wonder why it is needed. I showed you a related case a few weeks ago: how I built a development system for sales and marketing inside a commercial department, including a large block for working with Telegram. That brings us back to the website infrastructure and the problem of moving the site. My initial approach was to assume that I should choose a ready-made CMS. Some of our projects use Flexbe, others use Tilda, and still others rely on landing pages built on various CMS platforms inside other software.
I thought I might use a paid system, some SaaS product—Software as a Service, an online cloud platform—or perhaps connect an open-source solution. A second task was also important. When a person arrives on the site, I want to be able to ask them questions. I am known for liking questionnaires: I want to learn who people are, what they want to do, and what matters to them. We used this approach in education. When I was CEO of GeekBrains, a fairly large Russian-language provider of IT education, hundreds of thousands of people studied there simultaneously, including perhaps sixty or seventy thousand in long-term programs.
I always argued that when a person comes to receive a second education, it is important to learn at least the broad outline of where that person is in life and who they are, so that we understand whom we are serving. People would always ask me, “Why would anyone answer twenty, thirty, fifty, or sixty questions in a questionnaire?” My answer was: “If a person wants to change their life, it is probably reasonable for them to invest some time. When you apply for a driver’s license and receive a form with fifteen questions, you fill it out.
Even when buying an airline ticket, we answer questions. Yet everyone here assumes that a person should enter only a phone number and that conversion exists only when the form asks for nothing else.” I disagreed. With large-scale traffic—around one hundred thousand registrations per month—more than fifty percent of users completed our questionnaires. That was at mass-market scale. In real estate we use different questionnaires, and their conversion rate is often eighty to ninety percent.
The traffic volume is not one hundred thousand per month, but asking a person questions is still normal and natural. How else can you understand them? That is exactly why, in the episode about Claude Design and ChatGPT Voice, I explained why this matters.
Watch that episode; it explains the importance of asking questions. I needed a second module as well. The task was not merely to build website pages, but to be able to question a person—not only on the site, but later through Telegram, WhatsApp, SMS, a manager's phone call, or a voice robot. The channel did not matter. I needed the entire process to remain connected and manageable. I began searching again for open-source or paid questionnaire solutions. The more I studied the subject together with ChatGPT 5.5 Pro and Anthropic Opus 4.8 Max—I am naming the versions deliberately for those who care—the more clearly I saw that everything should be built in-house.
The system would repeatedly tell me: “Talk to your developers. They will spend a month on this, two months on that, and three months on something else.” When I replied, “I do not have developers. I will do it with Codex and Claude, and I do not want to look at a single line of code,” the answer changed: “In that case, the situation is different. You should abandon this CMS.” I am telling this story with a certain forward-looking understanding and vision of how the system should be built.
In some places I do not have to prove anything to anyone. If you go into your company or an outside business and propose this approach, you will encounter a huge number of constraints: how will it interact with the ERP, the CRM, the commercial department, marketing, marketers, and marketing directors? You will face many barriers that are difficult to cross. In my case I did not have to cross them. Where barriers do exist, I understand the problem and look for ways to act without confronting them directly.
Where the obstacle is too strong, I do not even propose the change. These systems matter to me directly because they either generate money, relate to my creative investment and time, or influence people. I want every person who encounters my videos, audio, articles, texts, meditations, paintings, classes, advice, recommendations, or anything else I create to receive the maximum value—to understand it and find what they need regardless of whether they are fifteen, thirty, forty, fifty, or seventy.
My audience is very diverse, just like the ToTheMoon audience. You can write to us about it. Viewers come from many different cities. We have not conducted a survey for a long time about which countries and cities people watch us from, and it is always interesting to learn. Naturally, our audience is Russian-speaking, and Russian-speaking people live all over the world—roughly three hundred million people. In the United States, for example, the Russian-speaking population is relatively small, about one million, unlike in some European countries.
I am not speaking only about Russia, Belarus, or Ukraine. I mean people around the world who know Russian, speak it, and are interested in watching this kind of video. As I worked through the CMS problem and received a design for the CMS, I set the next task. It was quite simple, and I understood the subject. I asked ChatGPT to prepare documentation for managing the site's front-end layout. I know how different platforms are built. This was not a simple website—building a simple website is now almost trivial.
A platform is different. I separated the front end and one of its most difficult areas: the quality of the implementation, the layout, and the division into reusable components. My projects are large and heavily loaded, so the system needs components instead of redrawing every button and every type of block from scratch. ChatGPT quickly wrote a technical specification. In fact, there were two specifications: one for the system that would create the pages, and another for the front-end implementation.
I gave them to Opus. Opus first passed the work to Fable and then to Opus 4.8 Max. When the Fable limits ran out, I did not yet know whether it was worth paying separately for additional tokens or what quality I would receive. I needed to work with it longer and see. Anthropic has recently shown some interesting ways to save tokens; we will discuss them separately, although the mechanism still looks complicated. In parallel, I asked ChatGPT—specifically Codex—to download my entire existing domain: every page, every form, the structure of every form, every detail, and all images.
Codex did it very quickly, efficiently, and with remarkable quality. I then handed the resulting structure back to Codex and said: “Implement this using these component requirements and these styles.” It did so very quickly. The site has perhaps sixty pages. Later I launched tasks involving three hundred or four hundred pages, and it handled those quickly as well. At the same time Anthropic was developing the internal management system. It began with several pages, brought them online quickly, and created a content management system inside the project.
That CMS gave me capabilities that no modern CMS had ever provided. It let me manage user sessions and cookies in great detail, see different forms on different landing pages, number and connect all of them, and automatically link contacts inside my commercial system. It immediately detected duplicates and applied the relevant international rules and standards.
It accounted for GDPR in Europe, where website-cookie requirements differ from those in the United States. Although American traffic was my initial priority, I designed the system for multiple traffic sources from the start. I also planned for load and speed: pages might need to be served from different servers around the world. You may not need that level of complexity, but I know that different countries have different page-loading conditions and different rules. I knew that search promotion mattered, including questions of indexing and non-indexing.
I immediately connected advertising systems as well. I knew I would need tracking counters, pixels, and various analytics tools. But when you examine tag management and analytics, you begin to see that you may not need any external system. You can install Google Analytics—or Yandex Metrica, which many people will use—and then Codex and Claude can program the entire data-collection layer for you: how people move, what they click, and everything else. Some viewers will hear this and say, “Sasha, that is impossibly complicated.” For some people it may be straightforward, but for most it will be difficult.
Let me give you another example involving front-end development and design. I asked the system to generate unique material for every U.S. state in which we operate, covering the aspects of my business relevant to that state, and to create and show me an individual page for each one. It produced the package. I sent that package into the front-end implementation process, then gave the result to Claude to deploy. Claude deployed everything, and an entire infrastructure appeared.
Imagine that you sell kettles and want a worldwide map of all kettles on your website. The example may sound strange, but the mechanism is the point. You ask ChatGPT to research the unique kettle market in every country, create all the pages and descriptions, and connect each page to conversion flows designed to sell your kettles. Today it can raise that infrastructure across the entire world almost instantly, with reasonably strong content. The result still depends on your own intelligence and maturity—how well you can connect, assemble, and manage everything—but the fact remains that websites can now be assembled almost instantly.
I am not really talking about CMS platforms today. I am asking whether you need a CMS at all. Do you need bot-management systems when you are planning something for your company, for the company where you work, or for yourself? I recently had another revealing case. I have many trips coming up—to different countries, sometimes with family, sometimes with my daughters, sometimes with my wife, sometimes alone, and sometimes meeting partners or friends. I return home in between and leave again.
There are many hotel and restaurant reservations inside that schedule.
Those hotels are paid with different cards and tracked in different systems. Some are prepaid, some cannot be changed, and in some cases I need the flight to be changeable even when the hotel is not. Some restaurant reservations are difficult to obtain. In some places seats have been selected; in others they have not. There are baggage rules and many other details. I finally asked ChatGPT to perform a task people have discussed for perhaps three years: analyzing email. I had never understood why I would want it to analyze my inbox, but over two weeks I had made so many travel purchases that I needed it to assemble everything.
It did. At one point it wrote: “Alexander, you booked a hotel in Honolulu, Hawaii, on the island of Oahu for Friday, but your flight arrives on that island on Saturday, and you leave another island, Maui, on Saturday rather than Friday. Are you sure you can make this restaurant reservation? Even if the dates are simply mixed up, the timing does not work: the restaurant is booked for 7:30 and your flight lands at 7:00. You cannot get there in time.” I opened the itinerary and wondered whether I had bought the ticket for the wrong date.
I had. I changed it, and the system saved me from several problems. The point is not the email case itself, but what booking and order-management systems will look like in the future and how all of this will be managed. Many different systems were involved. ChatGPT even matched each purchase to the card used, connected the data, and showed me the total cost—which, frankly, I did not want to see for all those hotels. The financial details matter because some cards provide valuable benefits: Platinum or Gold status with different companies, rental-car insurance, travel insurance, hotel-network privileges, and so on.
The system knows which hotel chains give me a particular status, where I receive major upgrades, and where I may be moved into a suite. An AI can therefore conduct a complete analysis and explain every opportunity. It does not merely say that checkout is normally at 11:00 a.m. It says: “With this card you can leave at 4:00 or 6:00 p.m. Instead of checking in at 4:00 p.m., your status may let you check in at 8:00 a.m. Here you could pay an additional amount for another option.” In another case it told me: “You are flying from Nice to Paris, arriving in Paris in the evening and leaving the next morning.
Are you sure you do not need a hotel? I cannot find a booking.” I knew I needed one; I simply had not booked it. The system inferred the missing step and identified it on its own. All of this brings us back to the systems that will exist in the future. If you are a startup founder, what are you building? If you invest in a project, which systems will still work? If you create and connect systems inside your business, are you sure that the configuration you are paying for today will remain correct over the next three to five years, or
are you spending your time on the wrong thing? If you work in a business and propose new ideas, are you implementing the right ones? To me, this is now the most important and fundamental question in the professional adoption of artificial intelligence. In ordinary personal life, many things can wait. I can go for days without using AI and nothing terrible will happen if I do not ask how to make a purchase or even how to treat a minor wound. Some things can wait or resolve themselves.
Professional decisions are different because they involve substantial investment. Even in healthcare, I am not usually investing money directly in the software. I choose a medical provider and insurance companies; I do not make a bet on the internal software they happen to use. But in your work and business, what are you betting on? Where are you spending a great deal of time while operating under a false assumption? Which areas do you not understand well enough to see the capabilities that already exist?
Everything has moved forward extraordinarily fast. I recently joked with my wife that a year ago she was a very advanced AI user, but today she no longer is, because systems such as ChatGPT and Claude have accelerated so dramatically. Understanding them now requires an investment of attention and time. That is one reason ToTheMoon exists. We will see one another in the next episodes and podcasts. Recommend the podcast and send it to other people. We want to develop it strongly and reach more subscribers and more people who are genuinely interested in this subject.
We are confident that a great many people are. See you soon.