How AI is changing the market in Russia in 2026, and what it means for a product manager
Banks, game development and the job market follow different rules, but the figures for 2025–2026 say the same thing: the shift is happening everywhere, faster than anyone is ready to admit. I collected the facts, checked the sources and worked out what follows for a product manager, and for me personally.
Natali Borisova · 1 September 2026 · 9 min read
I follow how AI enters different industries not out of curiosity, but because I work with the same tools every day. In short: in Russia, over 2025–2026, AI stopped being an experiment in banks, in game development and in job requirements. For a product manager, the speed of adoption is less interesting than what it changes in the product cycle: the distance between a hypothesis and a test of it has become shorter.
What is happening with AI in Russia right now
Banks already count the effect of AI in the hundreds of billions of rubles, the country’s largest tech company has released vibe coding as a mass product, and vacancies that require AI skills are growing faster than the job market as a whole. Here are four figures that set the scale. Below I unpack what each of them means for product work, not only for the market as a whole.
20%[1]
of financial-market organizations in Russia already use AI, according to the Bank of Russia
475+ billion ₽[2]
the effect of AI at Sber in 2025
10,777[7]
vacancies requiring AI skills on hh.ru at the start of 2026, +15% in a year
25 June 2026[10]
launch date of VibeCraft, Yandex’s vibe-coding service
What happened
In Russia, over 2025–2026, AI moved from pilots to a measurable effect across several industries at once.
Why it matters
These are no longer isolated experiments. Three independent markets are moving along the same adoption curve at the same time.
What changes for a PM
The question for a product manager is not whether to study AI, but at which stage of the product cycle it is actually useful to me.
AI in Russian banks
According to a consultative report from the Bank of Russia, one in five financial-market organizations already uses AI, and about another third plans to adopt it within three years[1]. From there the market splits into specific strategies.
Sber is scaling what already works: 700+ AI initiatives in 2025, an effect of more than 475 billion rubles, and a 2026 target of 550 billion[2][3]. VTB is at 15 billion rubles of effect so far, with a plan to reach 50 billion within two years[2]. T-Bank chose a third path. Instead of ready-made solutions, since 2024 it has been building its own models, T-lite and T-pro. The effect is several tens of billions of rubles for 2025, by the bank’s own estimate[8][9].
What interests me here is less the figures themselves than the fact that three banks chose three different answers to the same question: scale, catch up, or build your own. None of the answers is the only correct one. The choice usually depends on where the company already has an advantage.
Strategy 1
Sber — scaling
550 billion ₽
Target for the AI effect in 2026, after 475+ billion ₽ in 2025[2][3]
Strategy 2
VTB — catching up
15 → 50 billion ₽
Current effect, and the target over a two-year horizon[2]
Strategy 3
T-Bank — its own technology
T-lite / T-pro
Its own LLMs instead of ready-made solutions from outside[8]
What happened
Three large banks have publicly chosen three different strategies for adopting the same technology.
Why it matters
One technology does not produce one correct answer. The decision always depends on the company’s context.
What changes for a PM
Evaluating any AI tool is always a product and strategy decision, not simply a choice of technology.
AI in Russian game development
In game development the picture is less clear-cut than in banking, but the direction is the same. Industry experts call further adoption of AI in development and game design a key trend of 2026[4], trade publications write about an “AI boom”[5], and RBC Trends cites a forecast that neural networks will take more than half of the game production process within 5–10 years[6].
What happened
AI is taking hold even in a creative industry that was wary of it at the start.
Why it matters
If a technology sticks where resistance is highest, that is an infrastructure shift, not a fashion.
What changes for a PM
The limit of using AI is not the industry. It is which part of the process can be formalized at all.
How AI changes the product cycle
This is the main point. AI does not replace product skills. It shortens the distance between a hypothesis and a test of it. To show that in practice, not as a slogan, I broke an ordinary product cycle into stages and marked where AI actually shortens the path, and where the decision still belongs entirely to a person.
01
Problem
What is not working in the product right now
02 · AI
Research
Gather and structure what is already known
03
JTBD / CustDev
What the person is trying to do, and why they choose this way
04
Hypothesis
What we assume, and how we will know we were wrong
05
Prioritization
What to do first, and why
06 · AI
Vibe Coding / Prototype
Build a working version that can be shown
07
Usability Test
Show it to real people and watch what happens
08 · AI
Cohort Analysis
How different user groups behave after the change
09 · AI
Metrics
What the numbers show before and after
10
Decision
Whether to invest in a full build or not
AI clearly helps where the job is to gather, structure or prototype quickly: in research, in the first prototype, and in processing cohort and metric data. Where a live user interview, a prioritization decision or responsibility for a release is required, a product manager still does the work.
So this does not stay abstract, here is how the cycle above can look in practice. The scenario is hypothetical, but the sequence is a working one.
Imagine that a product’s registration conversion is falling. There is a problem, but the cause is not clear yet.
The product manager states a hypothesis: if registration is cut from five steps to three, conversion will rise.
AI helps review earlier interviews and data, assemble several interface options and quickly make a prototype. The product manager does not wait weeks in a development backlog. They build a working version themselves.
The product manager shows the prototype to users, collects feedback in a usability test, and after launch looks at conversion and how new cohorts behave.
If the hypothesis is not confirmed, the team does not spend several weeks fully building a solution that already looked weak at the test stage. The decision on whether to invest in development is made earlier, and on cheaper data.
Example: the Prototype stage in practice
A telling moment came on 25 June 2026: Yandex B2B Tech opened public access to VibeCraft, a service that assembles websites, CRM systems and working prototypes from a text description[10]. The company says outright that the service is also meant for founders and product managers who need to assemble a prototype quickly without a full development team[11]. It is based on Yandex AI Studio models and Yandex Cloud infrastructure, and the project code is stored in SourceCraft, from where it can be handed to developers[10].
I have been working in a similar way for more than a year, not through VibeCraft, but through my own stack of AI tools. General words about “working with AI” prove nothing, so here is the concrete version. I start with an md file where I write what the output should be, what the constraints are, and what counts as a finished result. Only after that does the task go to an AI agent in Cursor, one file at a time, without skipping steps. After each step the agent sends a summary of the changes, and I check it against the specification before allowing the next step. That comparison is usually where AI mistakes get caught. I am responsible for the logic, the specification and accepting each step. I bring in developers where heavy load, data security or complex integrations begin. I wrote about this process in a separate article, currently in Russian.
Before / now: what changes is not the competence, but the speed of moving between stages
Before
Now
The hypothesis, the prioritization and the decision to build remain product work in both cases. What changes is only how long the move between stages takes.
What happened
Vibe coding has become a mass product of large tech companies, not only my personal way of working.
Why it matters
This confirms that shortening the path from hypothesis to prototype is an infrastructure option, not one person’s trick.
What changes for a PM
A product manager can test some hypotheses before bringing in a development team. That does not replace the team. It changes the moment when the team is needed.
How AI is changing the Russian job market
According to hh.ru, the number of vacancies requiring AI skills grew from 9,378 in January–February 2025 to 10,777 in January–February 2026, a 15% rise in a year, and their share of all vacancies rose from 0.5% to 0.8%[7]. These requirements are gaining weight fastest in finance (+136%), security (+109%) and sales and customer service (+38%)[7]. At the same time, resumes mentioning AI were posted more than 200,000 times in 2025, and 55% of companies already believe that by 2028 AI literacy will be a basic competence for almost any profession[7]. In the second quarter of 2025, pay for specialists with AI skills was 34% higher than for colleagues without them: 100,700 versus 75,000 rubles[7].
Finance
+136%
Security
+109%
Sales and customer service
+38%
All vacancies requiring AI skills
+15%
Growth in vacancies requiring AI skills by industry, January–February 2025 to January–February 2026[7].
AI does not replace product skills. It shortens the distance between a hypothesis and a test of it.
What happened
Demand for AI skills is growing many times faster than the job market as a whole, and that is already visible in pay.
Why it matters
The premium is tangible now, not in theory and not in a forecast about the future.
What changes for a PM
Technical AI skills are becoming part of basic product literacy, not an option for “technical” product managers.
A map of the product manager and AI
If everything above is gathered into one diagram, the product manager’s role next to AI looks like this, without extra terms.
Research
AI helps gather and structure information about the market and users faster. Interviews and interpretation still belong to a person.
Product
Prioritization, product decisions and responsibility for the outcome are not delegated to AI.
AI
A tool for prototyping and processing data. It is part of the process, not a separate profession.
Validation
Testing a hypothesis with real users and metrics. AI speeds up the data work, but it does not replace the test.
Automation
Repeated operational tasks that can be handed to AI agents, freeing time for decisions.
AI skills that help a product manager
Not an exhaustive list of hard skills. Six areas that, in practice, turn out to be useful alongside product work.
Research
Phrase a request to AI so the result is structured, checkable information, not general phrases.
Hypothesis Testing
Turn a hypothesis into a testable statement, and know in advance which result would disprove it.
Prototyping
Build a working prototype with AI tools without waiting for a classic development cycle.
Data
Read metrics and cohort data well enough that AI processing does not become a black box.
Automation
Know which routine processes are worth handing to AI agents, and which are not.
Technical Literacy
A basic grasp of what AI models can and cannot do. Not programming, but a clear view of what you can actually get.
What this means for a product manager in 2026
If the map, the skills and the whole product cycle are gathered into a few practical conclusions, for me they sound like this.
A product manager should understand what AI tools can and cannot do as clearly as they understand product metrics. Otherwise someone else decides what gets automated.
A prototype can be built earlier in the process, before a development budget is approved, not after it.
Basic technical literacy is becoming part of product work, not a privilege of technical product managers.
Responsibility for interpreting the data stays with the product manager even when AI does the processing.
If everything said here is reduced to one split, it looks like this.
AI
Research
→ analysis
→ prototype
→ data processing
→ automation
are
needed
Product Manager
Problem
→ JTBD
→ hypothesis
→ priority
→ interpretation
→ decision
What I take from these changes
My path was never a straight line: sales → marketing → product → AI → vibe coding → automation. I now do product work on a retail team at VkusVill, seventeen of us there, and in parallel I build AI automations on the AI Empire project with a colleague.
This is not a resume for the sake of a resume. It is a way of thinking I use on every new task: quickly work out what is happening → state the task so it can be tested → assemble a working solution → measure the result. I use this cycle both for a product hypothesis and for a prompt to an AI agent. The difference is only the tool at the third step.
An example: when a client, Evgenia Postnikova, came with a request for a platform with an archetype test and a personal account, I did not assemble a development team. I wrote the specification myself and built the platform in Cursor, from the idea to production. That is the same cycle on a concrete task, not an abstraction from a resume.
Why it makes sense to start now
I do not think that in a year AI literacy will be a strict requirement in every product manager vacancy. I cannot confirm that forecast, and I do not want to present a wish as a fact. The reason to start now is simpler: with AI, the cycle from a hypothesis to a test of it is shorter, so each next hypothesis becomes cheaper in time, not only in budget. This is a rational choice, not a race after a trend. Learning the tool before it is urgent is usually cheaper than learning it under a deadline.
In short: my conclusions
- Banks, game development and the job market in Russia are moving in the same direction, almost in sync.
- AI does not replace product skills. It shortens the distance between a hypothesis and a test of it.
- In the product cycle, AI actually shortens research, the prototype and work with data. Hypotheses, prioritization and decisions stay with a person.
- What changes is not the product manager’s competence, but the speed of moving between stages of the cycle.
- It makes sense to start learning this now, not because of FOMO, but because learning ahead of time is cheaper than learning against a deadline.
Common questions
An approach where you describe the task in ordinary language, and an AI tool writes the code and assembles the interface, database and logic. It works well for fast prototypes and MVPs, but it does not replace an engineering team where heavy load, data security or complex integrations are required.
If you want to see how I work
Right now I am especially interested in products where you have to think about the user, the business and the technology at the same time. I like taking a hard problem apart, quickly assembling a working hypothesis and carrying it to a solution that can be tested in practice, with the same cycle described above, only shorter than before.
On borisova.one I collect projects, case studies and other material about product, AI, vibe coding and the growth of digital projects.
Sources
- 1.Bank of Russia, “Key conditions for the further development of artificial intelligence in the financial market” (primary source, 20 November 2025) — cbr.ru
- 2.Sber and VTB, the financial effect of AI in 2025 (from the “AI financier” report by the SberUniversity School of Finance) — finance.mail.ru
- 3.Sber, forecast of the AI effect for 2026 (statement by German Gref at the annual shareholders’ meeting) — interfax.ru
- 4.Tadviser, key trends in Russian game development for 2026 — tadviser.ru
- 5.ComNews, “A turning year. In 2026 the video game market expects an AI boom” — comnews.ru
- 6.RBC Trends, “Artificial intelligence in game development: prospects and trends” — trends.rbc.ru
- 7.hh.ru, “How AI is changing jobs, and who needs to learn to work with it” (primary source, 25 March 2026) — hh.ru
- 8.Tadviser, product card for T-lite (T-Bank’s own LLM) — tadviser.ru
- 9.Interfax, the economic effect of AI at T-Bank in 2025 (interview with Konstantin Markelov) — interfax.ru
- 10.Yandex, official announcement of the VibeCraft launch (primary source, 25 June 2026) — yandex.ru
- 11.rb.ru, VibeCraft is aimed at founders and product managers — rb.ru