How to Get the Most Out of AI in 2026: 10 Key Rules for Work and Life

Alexander analyzed every ToTheMoon episode, transcribed the whole body and asked Claude Code and, partly, ChatGPT to help formulate the thoughts he himself had been repeating for two years. There are twenty in total; they are published on the project's website with a description and the months in which each one came up.

A year or a year and a half ago everyone, OpenAI included, rushed to build artificial intelligence into the browser; Alexander said from the start that people would not work with it in that form. According to him, OpenAI recently wound the project down, and those who invested in the idea once again do not understand what will happen next.

A subscription costs about twenty dollars, and most people can afford it: it is not Netflix but an approach to developing yourself. ChatGPT drops even a paying user to a cheaper model, so the choice of the strongest one has to be rechecked; a plan for one or two hundred dollars pays off once automation in Codex or Claude Code begins.

If you limit yourself to checking the news, editing text or analyzing documents, you will not understand what modern artificial intelligence is and where it is heading. You have to make mistakes, see the model's mistakes and work out new scenarios — even where AI makes life harder at first.

Ilnar talks to people in China through Google Translate, but you can open ChatGPT Voice, ask it to translate both ways, explain complicated things without adding anything of its own and ask the other person again if something is unclear. The translator turns into a different business process.

The section with the thoughts on the website would once have required six months to a year and a team of developers, layout designers, marketers and editors. It also counts that the thought 'use AI all the time' came up in forty-one episodes over twelve months and seventy-two times over 2024–2026.

The main leaders are Anthropic and OpenAI, and you should work with them without bots, wrappers, aggregators or extra plugins: Alexander calls Cursor and Perplexity such middlemen. A token-counting plugin would have to be given all your systems and all your requests, while Claude and Codex themselves count tokens poorly — Codex has counted 49.8 billion for him.

On a rating image that, according to Alexander, Elon Musk more or less endorsed, Claude Opus 5.5 is in the SSS class, ChatGPT-6 Astra in the SS class, and Fable and another ChatGPT-6 model in the S class; he personally considers Fable 5.1 the stronger one. Google jokingly lands in a category of its own, and Chinese models, in his view, have already overtaken it in some respects.

There is no difference whether test results sit in Gmail, on Google Drive or in Telegram — or are uploaded to artificial intelligence: what you need to understand are the laws of the country where the service is located. In ChatGPT and at Anthropic, training on the user's data is on by default and can be switched off; Alexander does not trust Chinese models, where the checkbox guarantees nothing.

For fifteen years people argued about how Facebook, Google and Yandex learn from our clicks, and now people tell models everything themselves — from test results to tax returns. Privacy has long been relative: in a sane system nobody needs your health data, and if someone wants it, they will get it without you.

The fear of training on data rests on not knowing: training may not concern a specific person at all, and the value lies in the model learning from the data of competent people. People get strategies and solutions with the help of models — and forbid the models to learn from that best result.

More than a hundred financial questions about taxes, salaries, mortgages and loans were repeated more than ten thousand times, and the system was wrong in more than half of the cases. That is why answers have to be double-checked and the model asked to double-check itself: it can be wrong, understand differently or manipulate, and an answer can flip because of a single detail.

It does not matter who the expert is — Alexander Volchek, Sam Altman or Elon Musk: one can speak out of ignorance, hype or marketing, or simply be wrong, as OpenAI was with browsers. Alexander is going to OpenAI DevDay to look at opinions, because the company may have its own agenda.

It is worth uploading data of a certain range to AI — medical documents, the outcome of a visit to a psychotherapist, a teacher's advice — as a second opinion. It also works for clothes, a route or a restaurant, and the rule is worth taking on for a year: for decades we lived without such a tool and could not check a huge number of things.

Instead of micro-tasks inside your own process, artificial intelligence should be given the whole task: things are heading toward AGI, ASI and the technological singularity. With a legal question — tell the whole story from the start and attach the documents, rather than discussing one paragraph or one law.

The tenth rule, whose story began in the spring of 2026, is to automate your life in Codex or Claude Code, like the business processes of Alexander's wife with four children. In a company, a win is possible only when all employees use them: then Codex and Claude Code will become the central system, and there will be no CRM systems.