AI Is No Longer Just a Chatbot. It Is Replacing Workplace Software
Are OpenAI and Anthropic entering companies as consultants on deploying AI, or are they starting to compete with the makers of software themselves — and will third-party software survive at all once AI learns to build hardware and program for it?
OpenAI has declared itself a deployment company, bought the British Tamora with a hundred and fifty engineers and promises to invest four billion dollars, while on May 4 Anthropic, together with Blackstone, Hellman & Friedman and Goldman Sachs, created an AI services company — and the custom-development market, EPAM included, grew seriously worried. The host shows that a forward deployed engineer who builds the solution inside the client is no longer consulting but a path to software of one's own, and asks why talk to the IT staff when one can analyze all the source code, protocols and databases. Sam Altman had by then described AI as a utility in which a single super-valuable task may cost hundreds of millions of dollars of compute, and Elon Musk registered Macrohard and states the extreme pole — AI-native companies with their own operating system and their own compute. The host's closing thesis is that once AI learns to build hardware and program for it, no third-party software will remain, and in twelve months he proposes to check whether companies have started paying for AI rather than for seats.
What to watch for
Key takeaways
The host promises a simple picture: where old software stays and where artificial intelligence takes the value. For an entrepreneur it is a question of which product to build, for an office worker which tasks go to AI, for an investor whose money loses its power, for a developer who will write and maintain software. He proposes to check in twelve months whether companies have started paying genuinely for AI rather than for seats in programs.
Two years ago, when the channel started, the host was telling people in Silicon Valley that the model leaders were building more than a chat: for now they train on users, and once they have AGI they will generate software — so why a CRM system? A salesperson needs to know whom to call and whom to visit, not to look at a funnel interface; this is not "me and an agent" but something completely different. Hence the question whether an operating system will exist at all, whether OpenAI and similar companies will make their own OS and apps — and what happens to Apple, which does not let third-party developers into its infrastructure.
In March 2026 Sam Altman described AI not as a chatbot but as a utility — like electricity, water or gas. He first said it in 2024, and part of it was shown in the interview with Lex Fridman: a time will come when ten million dollars can be thrown at compute for a volume of tasks. In 2026 the wording got sharper: a single request can be cheap, while a single super-valuable task will require tens or hundreds of millions, and someday billions, of dollars of compute.
In recent weeks Alexander Volchek has been programming a lot with Codex and calls it phenomenal, although he has been in software for more than twenty years. But as soon as he asks the system for an operating system, internal messaging or task management, it runs out of computing power, even on the most expensive latest models without limits, and he has to work iteratively. His example: two hundred thousand dollars to analyze someone else's CRM and three hundred thousand to build it, and with the two billion Altman talks about — analyze Microsoft Windows and build Windows.
OpenAI announced its move from a model to a deployment company: by the official text, it should help organizations build and deploy AI systems for important daily work. In parallel it bought Tamora, a British company with a hundred and fifty developers and many partners, and promised four billion dollars of starting capital. These are forward deployed engineers: they do not sit in an OpenAI office but go inside the client, look at the data, permissions, chaos and legacy, and build the solution on the spot. The host's question is whether this is a consultant deploying third-party solutions or a company that will create its own software.
On May 4, about a week before OpenAI, Anthropic together with Blackstone, Hellman & Friedman and Goldman Sachs announced the creation of a new AI services company. It badly shook the custom-development market: the host's example is EPAM, founded in Belarus and serious even next to the world giants, and he relays what he heard from the market — companies are seriously worried. The new players are taking the very largest clients worldwide, and OpenAI has announced partnerships to serve, from a security standpoint, even European corporations.
Anthropic cites healthcare, where doctors spend hours on documentation, and the new company's team is meant to sit down with IT staff and doctors and build a tool around the existing processes. The host sees it as a patient of American medicine: the systems already record the conversation and produce a protocol, the steps are static for now but will soon become dynamic. His question is why talk to the IT staff at all when one can take the source code, protocols, reports and databases and analyze everything; hence the episode's main question — consultants or competitors of the software makers. The host's own bet is known: investing in infrastructure, from electricity and boards to the companies connected with it.
After his own Wikipedia, Elon Musk registered the trademark Macrohard — a counterpoint to Microsoft, "macro" and "hard" against "micro" and "soft". The market took it as a joke, but the host sees a direct path to the extreme pole Musk states: not agents inside a company but companies that are themselves artificial intelligence. Jensen Huang said a month ago that AI cannot yet build a company like NVIDIA, worth more than four trillion dollars, but can build something simpler — and the host asks what "simpler" means: a billion, twenty billion or a million dollars.
Musk speaks of a new era — new computers, new software, a new operating system whose kernel must be real artificial intelligence, not just AGI. NVIDIA is building such a layer together with computer makers, and Huang's position is: "I will sell the computer, sell the compute, so that you go on and do something with it." Musk's position: his own compute, his own AI, his own internet and his own operating system — down to the Tesla robot, which works offline and does external calculations on its own infrastructure. A third-party operating system is not needed in this logic.
The host's long-standing thesis, which many found populist: if artificial intelligence learns to create its own hardware and to program for it, no third-party software will remain — AI will create everything itself. The thesis grew out of his experience: twenty years ago he was implementing software at factories in Taiwan, where robots already moved along the walls and ceilings and few people were involved in producing complex processors. If AI can create infrastructure, it will program anything for it; when that happens is the question to viewers, and the host promises to check.
What this episode is about
A solo ToTheMoon episode in which Alexander Volchek assembles three pieces of news from the past month into one picture: where old software stays and where artificial intelligence takes the value. For an entrepreneur it is a question of which product to build, for an office worker which tasks will go to AI, for an investor whose money will lose its power, for a developer who will write and maintain software. In twelve months the host proposes to check whether companies start paying not for seats in programs but genuinely for artificial intelligence.
First comes the thesis the host says he was repeating two years ago to people in Silicon Valley: the model leaders are building more than a chat, they are training on users, and once they have AGI they will start generating software. Using CRM as the example, he explains why a salesperson needs to know whom to call rather than see a funnel, and why this is not "me and an agent" but something completely different. Hence the question whether an operating system will exist at all, whether OpenAI and similar companies will make their own OS and apps, and what happens in that case to Apple, which does not let third-party developers into its infrastructure.
Then come the words of Sam Altman, who in March 2026 described AI as a utility, like electricity, water and gas, and said that a single request can be cheap while a single super-valuable task will require tens or hundreds of millions, and someday billions, of dollars of compute. The host tests this on himself: in recent weeks he has been programming a lot with Codex and runs into the limits of compute even on the most expensive models — an operating system or a messaging system cannot be built in one go, he has to work iteratively. His thought experiment: spend two hundred thousand dollars analyzing someone else's CRM and three hundred thousand building it, and with two billion — analyze and build Windows.
The first piece of news is OpenAI moving to the notion of a deployment company: by the official text, it should help organizations build and deploy AI systems for important daily work. In parallel it bought the British Tamora with a hundred and fifty developers and many partners, and four billion dollars of starting capital has been promised. These engineers are forward deployed engineers: they do not sit in an OpenAI office but go inside the client, look at the data, permissions, chaos and legacy, and build solutions on the spot. The host's question: is this a consultant deploying third-party solutions, or a company that wants to understand everything and create its own software?
The second piece of news is Anthropic, which on May 4, a week before OpenAI, together with Blackstone, Hellman & Friedman and Goldman Sachs announced the creation of an AI services company. It shook the custom-development market: the host's example is EPAM, founded in Belarus, and he relays what he heard from the market — companies are seriously worried, because the new players are taking the very largest clients around the world and OpenAI promises to serve even European corporations. Using the healthcare example, where doctors spend hours on documentation, the host describes his own experience of American medicine — automatic recording of the conversation, static steps that will soon become dynamic — and asks why talk to the IT staff at all when one can take the source code, protocols, reports and databases and analyze everything. His own bet, he says, is known: investing in infrastructure, from electricity and boards to the companies connected with it.
The third story is Elon Musk, whom everyone at first took as a joke: first his own Wikipedia, then the trademark Macrohard as a counterpoint to Microsoft. Musk states the extreme pole: not agents inside a company but AI-native software companies that are themselves artificial intelligence. Jensen Huang, according to the host, said a month ago that AI cannot yet build a company like NVIDIA, worth more than four trillion dollars, but can build something simpler — and the host asks what "simpler" means: a billion, twenty billion or a million dollars.
From there the conversation turns to a new operating system whose kernel must be real artificial intelligence. NVIDIA is building such a layer together with computer makers, and Huang's position is to sell the computer and the compute so that others do something on them; Musk's position is his own compute, his own AI, his own internet and his own operating system, down to a Tesla robot working offline. In the finale the host returns to his long-standing thesis, which grew out of implementing software at factories in Taiwan twenty years ago: if AI learns to build hardware and program for it, there will be no third-party software — and he invites viewers to answer when that will happen.
The episode is useful because it ties three scattered pieces of news into one line: OpenAI and Anthropic are buying and building teams that sit down inside the client, while Musk announces companies that are themselves AI. The host does not hide that he is betting on his own scenario — software does not die, it becomes invisible — but he also sets a verifiable deadline: twelve months to see whether companies have started paying for AI rather than for seats. He names the main constraint himself: today even he lacks the compute to build a large system in one go, so the question comes down to compute and to who owns it.
Episode transcript
The episode is in Russian; below is an English reading guide to the transcript (the full EN transcript is a machine translation). Voice matching applied to 31 segments: 0 identified, 0 mixed, 31 probable, and 0 unresolved.
Read transcript on a separate page
Loading…