Hi everyone, this is the ToTheMoon channel, Alexander Volchek. I want to discuss the following topic with you today, very briefly. Your opinion matters, and so does the experience of a great many people. Today I'll share some aspects of how working together with artificial intelligence looks for an ordinary person. So there's a chat everyone is used to working with, and most people in the world, hundreds of millions of them, use it daily. They ask it some kind of questions, for instance, and usually stop there. Clearly there's a next category of people, and there are plenty of them on this channel, who start building various solutions, products, software solutions with the help of OpenAI, for example Codex, or Claude Code. And today we'll be sitting roughly between those two.
The question that is fundamentally on the table right now: do you really need to build, say, a mobile app for work? Do you really need to build some system with an interface, to build dashboards, reports? Or is what exists today enough — the command line, so to speak, this chat line, so to speak. I put forward a hypothesis more than once, three years ago already, that with systems like ChatGPT around, various specialised solutions would obviously start disappearing. And as an example I always went as far as saying it would be CRM solutions, ones like Salesforce, or on the Russian-speaking market, ones like amoCRM.
Such systems should gradually go away as a layer. There are simpler layers too, of course. Systems used for generating PDF files, say, for visualising reports, like Power BI, for building some kind of marketing analytics. Systems like that will go even faster. I absolutely don't want to say that Salesforce disappears tomorrow, or that amoCRM disappears tomorrow. It won't. But from the point of view of how efficient they are to use, why will they be going away overall? Because there is — well, in my case at least there is — a fairly strong understanding that if a person needs to carry out some action, then the optimised path of that action, in many cases, not all of course, but in many cases it can be reached through the ChatGPT command line.
Or through a great many other interfaces that will be needed. And with the CRM system in particular, I gave the example: why should a sales manager open a CRM to find out whom to call, if you can open ChatGPT and type, whom should I call right now, which fifteen people, right? And the telephony integration will be right there, because that's elementary, right? Or he opens Telegram, say, or opens WhatsApp and asks whom to call. The system tells him whom to call. Again, using the architecture of Codex or Claude, say, of systems like ChatGPT.
Because these systems will be integrated with all of that anyway. A simple example, the closest one of that kind of development — that's how OpenClove appeared, as a layer between the strong models, ones like OpenAI, Anthropic or Google Gemini, and said that we let you quickly build a solution so that a person could receive some messages in WhatsApp, for instance. Well, or receive messages in messengers. I said right then that there is an enormous number of business processes that don't need WhatsApp at all, right?
And it's odd to keep that as a layer, if I have ChatGPT — ChatGPT will perform that function of telling me something somewhere. And essentially OpenClove, through various integrations and mergers, vanished the way a hype story does, and what came up immediately instead was working in OpenAI Codex or in Claude Code. So, which side are you on right now? Let's write it down. There should probably be a big poll here, by rights. Which side are you on? Are you for there being software that gets launched automatically out of systems like ChatGPT or Claude?
Clearly the very concept of ChatGPT together with Codex or Claude Code, Claude Design, coworks, all of that — it will transform somehow and turn into something else. Perhaps some equipment will appear, some hardware devices, some audio recorders we'll be able to speak into. Or clearly these systems will be embedded somewhere else as well. Incidentally, the model of embedding these systems, of integrating them in different places, hasn't worked out especially well. As, in particular, they repeatedly tried to integrate ChatGPT with Alexa. It isn't, well, a widespread case, right, for the Amazon speaker, which is used fairly widely and at scale, to have a link to OpenAI, and for that to work brilliantly. So at some point you think, why do I need this, if, again, everyone carries a phone around with them.
A person can take out the phone and ask it something. Right, let's go through several cases.
Take even this example: I run more than ten projects in parallel in Codex, and in Claude. Clearly I have hundreds of business processes organised with the help of artificial intelligence. I mean some kinds of solutions here. There are simple ones, like a system that informs me about a certain part of the financial market, since I invest quite a lot in what is more of an AI-tech sector. OpenAI pays companies, those companies pay the ones they invest in. Roughly that kind of spectrum.
Well, clearly they have an enormous number of different contracts directly. I need to receive certain information on the financial market. So an integration with Telegram is set up on the server, an integration with various API services, with finance, is set up. And certain notifications reach me, alerts in a Telegram bot, right, or reach some people connected to me who need that alerting. Do I have an interface for managing this process? I don't. If I need to update some portfolios, or I need to get some reporting or some information — I either do it through a trained Telegram, which works directly with my API system, so to speak, or I open Codex and write in the Telegram project.
So, make me such-and-such a report, or get me this information. Do I need an interface here? I don't. It isn't even clear what this interface could be needed for at all, even in theory, what it could be needed for. Let's complicate the system, right, go further. I have, for example, a system I built so that it integrates with cameras, recognises the movement of objects — people, animals,
cars, plates, faces — at various set times, and controls some of my infrastructure, the gate for instance, and some other things around my house. Yes, it's a kind of project where I can connect various smart-home systems and control them. Do I build interfaces for this? I don't need interfaces; sometimes I can say, "Give me some report or show me a page on the web" so that I can quickly look through some images, for instance. But on the whole I don't even need that web interface.
I have it by default anyway through various open-source face-recognition solutions. And do you know what for? Because there's a certain habit about how these systems are supposed to work. Although essentially, with the next system as an example, I'll show that building an interface is practically unrealistic, very hard. For instance, I have a great many projects tied to areas where there's a lot of varied content, and I need to analyse people's voices.
And I have three projects running, well, two fundamentally on this subject, running completely independently, which analyse an enormous number of different calls, which analyse an enormous number of videos, which analyse an enormous number of audio messages and then turn that into voice and so on. And what did I run into? Into designing an interface convenient for, say, validating the right voices. This voice is Sasha's or Masha's, this one is Katya's or Petya's, right? In the course of the work that's very hard.
That is, you have to work for months and describe how it should even work visually. And most importantly, I wouldn't want to build one simple interface; I'd want to do whatever I want at any moment. For instance, I'm sitting there and I say, "You wrote that our voice match is ninety-five per cent." I say, "One hundred." It says, "Then you need to check six hundred cases." I say, "I won't check six hundred, but I'll check about forty by hand." And it starts giving them to me as short phrases in Codex, as short little audio excerpts. I listen to them and say, "Yes, that's this person, this one is this, that one is that." But that interface, essentially, runs out of the chat. That is, I didn't build any special software for it.
It runs for me straight out of the chat. Remember there was one case where I very quickly wrote software for the Mac so that it would read from a sensor —
I have a Polaris sensor, I think, or whatever it's called, and that sensor is for professional measurement of everything to do with the heart, or semi-professional measurement. Twenty-four seven. And I really disliked having to launch one piece of software and then launch Matlab. And essentially I was interested in getting the data from that sensor directly and passing it to ChatGPT. I wrote a program to start the sensor. Now I realise I don't need the program. And there were constant problems with the program. It ate the wrong sort of memory on the Mac, the synchronisation was somehow wrong.
I realise that all my synchronisation runs through Codex. That is, I calmly tell Codex, "Start my sensor for the night." It calmly starts it, since it's nearby, and that's it. In the morning I tell it, "Switch it off, download the data." It downloaded the data for me and I immediately tell it, "Build me a report, or compare my readings, or do everything I want. For instance, go into my account with my current provider and send this data to my doctor. Or compare this data with three years of cardiogram data. Or find me in my mail, in my notes, in all my files, everything I've ever had on how my heart works, and make a variety of it." My system is fairly healthy as far as service goes. But I was curious, I was studying things to do with sleep, to do with various detailed indicators, those modern indicators of health analysis, right? So the software is sort of not needed.
The next piece of software, for example, was a solution I spoke about recently, and that was exactly one of the stories where I said, right, farewell to financial-control
systems and various accounting ones. And it turned out there's the American system QuickBooks, and within the American system QuickBooks I integrated with it over the API, and so Codex downloaded absolutely all the data. Then it synchronised with my statements from the banks. That's with my bank statements. Clearly I didn't give it access to the banks. I have no interest in that, but I did give it full access to the data, to all the PDF exports from the banks, so to speak, and it built me reporting far better and more convenient than QuickBooks does and than any system would do for me.
Because I loaded all the tax filings in there, I loaded in all my data for the current year. This was done within one business entity, one corporation. And I also liked one thing about the system: for instance, even — someone was saying recently, "But what about QuickBooks? QuickBooks stores receipts." Well, and what do I need that for? I tell Codex, "Listen, go into Expedia, download, for example, my receipts from this trip." It goes in, sees absolutely everything and says, "Here, we can see you paid for a spa, so to speak; it should come out of this business's expenses, right?" Well, it clearly understands that a spa, for instance, isn't business. Or, here you travelled to a particular city — did you definitely go there for work, or do these figures, do these costs have to be written off as your personal expenses.
Not one system will do that as easily and conveniently. Or it says to me, "Now here I don't understand. Here three hundred and seventy dollars were spent on something. What is this?" I say, "I don't remember what contractor that was. Find how such a contractor is usually labelled on bank statements." Everyone it didn't know — across twenty or thirty transactions it found the people, found the companies. It immediately worked out what the company was. It immediately worked out what they sell. It suggested whether this relates to my business or not, what the details are and so on. So it did absolutely everything. Do I need an interface here? I don't.
I don't need an interface at all. Do I need an interface? Today I have two large site-management systems running, and the site that's running is, incidentally,
ToTheMoon, but it sits within my Besolid projects, besolid.com, right? And, well, the domains are down below, by the way. And the ToTheMoon site is there. So. And then there's a second story, where I recently did — you know these approaches I had, of building a CMS, of building sites — I did a CMS update there for all our sites for the real-estate company. And as an example, actually, there's a whole range of different projects there, but these are the big ones, fairly big loaded sites, you could even call them semi-portals or portal systems. And whereas in real estate I put a great deal of time into building a manageable CMS system, on my own site, the more different things I shipped, the more I understood that I don't need a CMS system. For instance, it used to be necessary to write tags and various parameters into every page by hand so that your site would come out better in search promotion, right? Now, first of all, it's odd to do that by hand, and secondly, the algorithms are strong — Fable, say, or 5.6 Sol will describe far better than I can everything that's going on from the point of view of, well, SEO. They'll re-check all the texts, in places they'll say you need to add text, in places they'll even generate the text and write it in.
They'll get all the settings right, index all the images properly — in short, they'll get an enormous number of settings right. That doesn't mean it's all done perfectly somewhere, but it's clearly a story that's going to develop. I think that the ToTheMoon project, or the Besolid projects, will in about three months have resources, in terms of sites and portals, that it would have been impossible to build by any means, even by a serious group of people. Because what's been done at ToTheMoon in terms of semantics — and a lot has now been done at Besolid too in terms of semantics — it's just that a lot of it hasn't been laid out or programmed there yet.
What they have done is, well, very serious, unbelievably serious, right? And you immediately ask yourself the question: so is an interface needed at all then?
And here's one of the topics that Ilnar and I discussed, I think, in the Sunday episode — the topic that right now there's been another partnership integration between OpenAI and a mobile-app development company, one like Wabble, worth more than ten billion dollars, which lets you build mobile apps. What code do they use to build a mobile app? Of course, Anthropic or OpenAI. And the question is: why do I need third-party pads so that some solution gets built for me with a limitation?
Incidentally, let me add a preamble right away. Obviously, look, obviously there's a range of things where an interface is absolutely necessary. I am in no way removing the case here in which an interface is not needed at all, anywhere. But I understand — I have a lot of experience building platforms, an enormous amount of experience building various platforms, and I really do love it, I'm a fanatic about it. I can say there's a habit, there's even a habit when it comes to managing the leads that come in from websites, say, or the production requests that come in, or service requests, or ticketing systems, support-desk requests, requests coming in. There's a certain habit that you need a list, that you need some visualisation of the list.
But is it really needed, or do you need, again, this terminal where you set the task, you say, "Show me particular data," and it shows it to me. There's a question somewhere in there: what's the speed of getting that data, what's the volume of data processing. But the more different projects I do, the more I understand that I'm better off without an interface. Because an interface is extra load, extra servers, it's always problems with visualisation, problems with layout, problems with the speed of operation.
or has the right infrastructure set up. But if we're talking about the world at scale — I'm not talking about big corporations; big corporations will sit for a very long time on their internal software, and they'll hit impossible problems rewriting it, impossible problems. Even before artificial intelligence, all of them had enormous problems rewriting software. It's like taking — there's a company in the Russian Federation, Russian Railways. And if you look carefully at how they issue tickets there, I don't know about now, but there they always issued tickets, and the software was like a terminal, like the old, what were they called, text editors of that kind.
You practically had to type text in order to issue a ticket. That is, that's how they worked. They ran a command, as it were. After all, that's how it originally was: you ran a command, as it were. Clearly you can't run just any command there, write just any phrases, right? And an enormous number of companies never managed to rewrite that software, and it stayed in the infrastructure of dozens, hundreds of different solutions — why, only the day before yesterday I was on Fable connecting a piece of the site, www.besolid.com in fact, to the main domain. And it says to me, "You need to connect Google Search Console." And I sit there thinking, what is it for at all? What is it for, at bottom? I connected it there through Google Cloud, connected all those things, but what is it for?
Well, counters are fine, we'll put counters in. And even the counters that are there — they're the ones that are more likely needed. That is, Google Search Console was more likely needed, and essentially that was the task — in this case it was Claude, yes, it was Claude. It said to connect it so that the SEO would be solid, so that the search promotion would be solid. But I understand that there won't be any search promotion either. And when you understand that there won't be search promotion, that the system will work differently, you of course ask yourself this big question: so will it be the command line after all, or will it be an enormous amount of software?
What do you think? Write here — I'd really like to hear and understand your opinion on this. Do give examples of where you see that you can't migrate software, and where you can migrate software. As I'd say — I can say that my paradigm is of course changing very strongly, though, again, I have no fundamental paradigm of software. You can see that in all our videos. And as of today ToTheMoon already has more than a hundred and fifty videos. Bye everyone. See you in the new episode.