Who is Alexander Volchek
He tells the story of his path over the course of the episodes. Starting in 2003, after he became fascinated with data normalization, Alexander worked on enterprise automation — ERP, CRM and other systems. He used to build them with teams of people; now, he says, he can put them together himself, and sometimes that pays off better. He also brings AI into the companies where he becomes a shareholder or a board member: beforehand, he loads tens of thousands of pages, calls and tickets into a model to understand the business from the inside.
ToTheMoon comes in different formats. Alexander often records his solo episodes on the road between Los Angeles, Silicon Valley and Santa Barbara; in the new digest format, thirty-five news stories of the week run back to back, each with his assessment; and there are long conversations with Anna Volchek, Ilnar Shafigullin and Max Grigoriev. His co-hosts often argue with him — for example, over whether Anthropic will have enough money for its commitments and where the main money in AI lies: in the enterprise or with ordinary people.
The ideas he keeps coming back to
Twenty ideas that Alexander repeats from episode to episode are collected in the “Trends” section on besolid.com — drawn from every ToTheMoon transcript since April 2024, with quotes and timecodes. Many of them are practical. Pay for AI and use a strong model: a friend who complained about a “dumb” Codex had simply been running it on the lightest one. Keep one main system and a backup: a third and a fourth, in his experience, are hard to keep up with. Judge models by hundreds of your own tasks, not by leaderboards.
Other ideas are about how work is organized. Bolting AI onto old processes, Alexander believes, is a mistake: in the days of ERP and CRM, processes were automated as they were, while strong systems such as Airbnb or Uber were invented from scratch. Hence his advice to finance people to keep their books as if QuickBooks and 1C had already disappeared, and the conclusion he drew from the November news: all software will have to be rewritten, even software built with AI. He considers the hype around “agents” mostly marketing.
There are forecasts too. Strong AI, in his view, will become more expensive: a hundred-dollar ChatGPT Pro subscription today, by his estimate, delivers a thousand times more value than it costs, but that kind of generosity will not stretch to a billion people. And real AI, as Alexander sees it, should not close the task at all: it comes back to the person an hour or a day later with new conclusions of its own — which is what Dots already does.
How he works with AI himself
Many of his tips grew out of his own tasks. Codex went through several months of hotel bookings, found where breakfast credits had not been applied, and pressed the hotels for refunds on its own until part of the money came back. It also connected, without any API, to the Amazon cloud where Alexander's more than twenty cameras store their video. For a server computer bought on a personal card, a short remark was enough: Codex found the purchase and the receipts in his email and accounts by itself.
80% of his email goes through Dots; the assistant sorts out the school newsletter on its own, and bank notifications no longer land in his inbox. For development Alexander uses Claude Code on Opus 5.5 and Fable, and entrusts the rest of his life to ChatGPT, Codex and Dots. In his projects, AI handles user requests: it scans attachments with an antivirus, fixes bugs and replies to people, and brings ideas to him for a decision. He counts the tokens he has used in tens of billions.
He talks about failures too. With a $350 limit in AWS, while Claude was converting video on the servers, Amazon showed $37,700 in charges after ten minutes; it later turned out that Amazon itself had made a mistake, and the amount changed to $112. Dots once sent an email without approval, against a rule, and Codex on a Mac asked for access it did not need. Hence his advice: watch the limits in every system you have entrusted with something, and do not treat one model as a reliable check on another.
Where to start listening
For a first introduction, try episode 183: Alexander tests Codex, Claude Code, Google Antigravity and Cursor not on code but on everyday errands, and at the end lists seven criteria for comparing them. Next comes episode 191, on how to turn Codex or Claude Code into a personal financier with three rules. For executives — episode 190, on his experience of enterprise automation and why the systems that win are the ones invented anew.
The big picture comes from episode 176 with twelve trends — on money, power and the future of humans — and from conversation 182 with Max Grigoriev about what counts as intelligence and why AGI has become a marketing word. The recent solo episode 195 on Fable 5.2 also has an uncomfortable question: Alexander asks listeners to rate their knowledge of AI on a scale from zero to a hundred and expects that many, after the episodes on trends, will give themselves a zero. Episodes come out on YouTube, and transcripts on besolid.com.