How Do You Earn More With AI When the Competition Only Grows?
Where is the money in AI really, now that being able to produce text, analytics or code fast is no longer rare — and what separates those whose income it raised from those who merely got faster?
The reader gets a sober picture of the market through a client's eyes — why hiring has not got easier even though there should be more people available — and a concrete path instead of generalities: push everyday chat use to a few dozen requests a day, move half your tasks out of the chat into Codex and Claude Code, cut the number of third-party tools, and treat the subscription as an investment rather than a cost.
What to watch for
Key takeaways
The paradox the episode opens on: what used to take hours or a whole team, one person now does many times faster — and earning has not got easier. Clients are still just as hard to find, competition is growing, and the bar for a specialist keeps rising.
Even when AI lets you go many times faster — pull data quickly, produce an analytical report, write a text — everything runs into the same thing it ran into before. The capacity to do the work has grown; the capacity to sell it has not.
The author looks from the other side of the table — he owns companies, sits on boards, hires constantly. In three years the ease of finding contractors has not risen, though there should be markedly more people able to do marketing, sales, analytics, editing and code. He sees no correlation, and neither do the businesses he talks to.
The explanation is simple: the person wanting to earn with technology wants more. The person commissioning the work wants to pay less — or, if willing to pay the same, to get higher effectiveness and lower hiring risk. Hence the absence of the promised relief.
In his own hiring the author records not a higher bar but a changed approach to whom he wants: a person with a different spread of professional qualities. Because of that, hiring has become harder rather than easier.
The episode's key claim. An artist setting out now to open a business and learn marketing and sales has not had it made easier — it got harder: the more people who know something, the harder it is to sell. The one who was already in the market wins.
Opportunity number one is to push ordinary chats, ChatGPT and Claude, up to a few dozen requests a day: thirty-five, fifty, seventy, on any subject. The author compares it to when you had to know basic Excel or Word functions: not an exotic skill but the norm of daily work.
The next step is to start doing half your tasks and more not in a plain chat but yourself, in Codex or Claude Code. The author gives a live example from the Sunday podcast: Tatyana has not moved her simple projects into Codex, and that is exactly the line in question.
The practical half of the same step: whatever you open a separate service for — speech recognition, assembling a PDF or PowerPoint deck in Figma or Miro — try doing inside the system itself. The aim is for more business processes to run directly in Codex or Claude Code.
The author answers the charge of fixation: Codex and Claude Code are named not out of habit but because they have capabilities Gemini, Grok and DeepSeek do not have today. He puts the principle more broadly: use the best rather than settle for leftovers — with a caveat about company, location and legal restrictions.
The author only partly accepts the talk of businesses now opening for free. If twenty dollars for an entry-level paid tier is out of reach, then discussing a five-to-seven-hundred rise in income makes no sense. He calls the thinking inverted: first we raise the income, then we decide whether to pay twenty dollars.
Taking courses to learn Codex or Claude Code the author considers optional: thirty to fifty requests a day on your own real subjects will teach you faster, and you can ask the system itself how to work in it. He does not reject courses but judges them by the investment: a thousand dollars a year against a hundred-dollar subscription — and usually several times over in the subscription's favour.
A separate thought: neither the viewer nor the author himself can learn one business process in Codex or Claude ahead of time. What a year ago needed a strong specialist is elementary today, and effectiveness will keep rising. The task is not a specific action but understanding how the system works and being unafraid to go into it.
The author's conclusion, drawn from his consulting practice and work with companies: the only piece of advice that can honestly be given is to use AI in your own work. That is the only way to become a real professional at it.
What this episode is about
A solo episode by Alexander Volchek about where the money in AI actually is. The starting point is a paradox: what used to take hours or a whole team, one person now does many times faster — and earning has not got any easier.
The first block is the view from the other side of the table. The author looks at the market as a client: he owns companies, sits on boards, hires constantly. Over the past three years finding people has not become easier — even though there should be markedly more of them able to do marketing, analytics, text and code. The reason is that the two sides move towards each other: the contractor wants to earn more, the client wants to pay less or get higher effectiveness and lower risk. His own requirements have not so much risen as changed: he wants a person with a different spread of qualities, and hiring has got harder, not easier.
Hence the first conclusion, stated plainly: AI helps not beginners first of all but those already in the market. An artist setting out now to open a business and learn sales has not had it made easier — it has got harder, because marketing became harder as competition grew. The main problem is what it was before AI: finding a client.
The second block is what to do about it. Step one: ordinary chats have to become an everyday norm — thirty, fifty, seventy requests a day on any subject, from health and food to work questions — the way basic Excel or Word once had to be known. Step two, the main one: move half to two-thirds of your tasks out of the chat into Codex and Claude Code and do them there yourself, cutting the number of third-party tools — not a separate speech-recognition service, not Figma or Miro for a deck, but the same thing inside the system. The author explains why he names these two specifically: they have capabilities Gemini, Grok and DeepSeek do not — and gives an example from the Sunday podcast, where Tatyana has not moved her simple projects into Codex.
The third block is money and training. He accepts talk of businesses now opening for free only in part: if twenty dollars for an entry-level paid tier is out of reach, then a conversation about raising income by five hundred makes no sense. He calls the thinking inverted: people want to raise their income first and only then decide whether to pay twenty dollars. He does not reject courses but judges them by the investment: for a thousand dollars a year, take the hundred-dollar subscription instead — the effectiveness is usually several times better, though not guaranteed. What teaches you is not a course but daily practice on real tasks, all the more so because you cannot prepare in advance: what a year ago took a strong specialist in Codex is elementary today.
The episode honestly separates two things that conversations about earning usually blur: speed and income. Speed is available to everyone today and has therefore stopped being an advantage; income still comes down to the client, and finding clients has got harder rather than easier — competition grew faster than the capabilities did. Hence the author's prescription is not about a tool but about depth: stop looking for one more way to go faster, cut your tools down to two or three, and go deep enough in them to do yourself what you used to hire out. His caveat stays with the reader: there are no guarantees, everyone's circumstances differ, and where exactly to build effectiveness is a decision each person makes alone.
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 32 segments: 32 identified, 0 mixed, 0 probable, and 0 unresolved.
Loading…