The main mistake of tech-driven businesses today is believing that technology adoption is complete once a service is selected, a pilot is launched, or a new department is created. In reality, that is precisely when the transformation begins. Technology does not change just a single operation; it transforms the connections between tasks, people, data, quality criteria, and accountability for results. The most striking trend, however, is that most users of accessible AI (large language models, chatbots) are not monitored by their employers in terms of how and for what purposes they use it at work. In other words, even businesses that do not officially use AI are in fact using it at the level of routine employee actions — employees turn to AI haphazardly, without corporate guidelines or an understanding of risk areas.
The first common mistake is starting with the capabilities of the technology rather than with how the work is organized. Asking “Where can we apply AI?” almost inevitably leads to a collection of disconnected pilots. It is far more productive to dissect a specific process: where delays occur, at what stage quality is lost, what decision a human makes, and what data they use for it. After such analysis, it becomes clear where AI genuinely creates value and where it merely adds yet another interface. This approach is embedded in the strongest current AI-in-business programs (Oxford, MIT, and others).
The second mistake is automating entire professions and departments without breaking them down into individual tasks. AI does not possess a uniform level of intelligence; it can perform one task flawlessly while failing at a superficially similar one. Researchers refer to this as “jagged technological frontier.” Therefore, one cannot make blanket statements that AI can analyze, replace a marketer, or prepare management decisions. Its capabilities must be tested against specific data, operations, and quality criteria. Designing these actions necessarily includes a pilot stage, but what is even more important — and this gradually fosters systemic thinking and internal understanding within a company’s environment — is determining what we can delegate to AI and what we cannot. And if we do delegate, what frameworks (metrics) do we use?
The third mistake is scaling the technology before the organization has defined the new role of the human. AI can prepare forecasts, options, or recommendations, but professional judgment, choice, and accountability remain human functions. If it is not established who verifies the source data, spots gaps, makes the final decision, and takes responsibility for consequences, accelerating work may simultaneously mean accelerating errors. This point actually debunks some of the fears that AI will leave us without jobs. It will not displace those who continue to build competence in their field. AI can optimize our work in a number of areas, but control points and value will not emerge without a human expert.
These problems can be avoided by following a sequence: first, study the actual process; then conduct a limited experiment, measuring not only speed but also quality, errors, and the burden on people; and only after that, scale the solution. At the same time, it is necessary to develop employee competencies: the ability to formulate tasks, verify AI outputs, see alternatives, and explain their own decisions. General literacy and digital resilience at the individual level, for each employee, are now among the most critical challenges for society. We need to gradually align our collective understanding of what this technology is that can so profoundly affect society — and not only in the corporate environment.
Russian tech-driven businesses are currently not at a point of decline, but rather in a complex and promising transition from demonstration projects to the routine use of technology. According to the Bank of Russia, one in seven enterprises already uses AI, although the maturity of practices varies widely. This is a natural stage: infrastructure is developing faster than management models and skills.
The positive outlook is that the market is beginning to mature. Companies are increasingly asking not which neural network to buy but how to restructure their work and achieve verifiable results. The competitive advantage will go not to those with the most tools, but to those who learn fastest to turn technological capabilities into sustainable organizational competence: experimenting, preserving professional judgment, developing people, and being accountable for outcomes.
I see a particular opportunity here for Russian businesses. Russia was not the first country to mass-produce large language models and experiment with generative AI. But entering later can be an advantage: we already have access to a generation of significantly more productive systems, as well as the accumulated global knowledge about their errors, limitations, and impact on human work.
Russian business has a strong tradition of fundamental training, engineering thinking, and the ability to operate under constraints. However, these qualities alone do not guarantee an advantage. It must be built — starting AI adoption not with a race for tools, but with developing mindset, professional judgment, and a new culture of accountability.
Perhaps the mission of Russian tech-driven business right now is not to repeat someone else’s path from technological euphoria to the realization of risks, but to enter the AI era in a more mature way from the outset — not only creating its own models and platforms, but simultaneously designing new collaborative workflows between humans and technology. If we start with a mindset, we have a chance to be among the leaders not by the number of early experiments, but by the quality of real-world application.
Maria Lopukhina, Director of the 1331 AI Studio, expert in AI in education, management and corporate environment