Olga Chernokoz: Standards for AI—where rules end and restricting competition begins

A recent decision to create, under the auspices of the Ministry of Digital Development, a Council on generative software development raises a broader question: what exactly should the state standardize in the rapidly changing field of artificial intelligence—requirements for the outcome or the very ways of achieving it?

Sber has proposed creating an industry platform for the development of generative software development. It is assumed that the scientific community, customers, and developers of solutions based on artificial intelligence will take part in its work.

Among the stated objectives are the creation of a unified methodological framework, approaches to assessing the economic effect of AI adoption, and the efficiency of computing resource usage. In the future, the results of this work may become the basis for national standards in the area of agent-based development.

At first glance, the initiative is вполне logical. Generative AI really is changing software development. AI agents are already able to write and analyze code, conduct testing, search for errors, and perform some of the tasks that previously were solved directly by programmers.

An objective problem of measuring outcomes is also emerging. If one company claims that the use of AI increased developer productivity by 20%, and another by 50%, it is necessary to understand what exactly was measured, whether the costs of computing resources were taken into account, how product quality changed, and whether an economic effect was actually achieved.

Unified and transparent approaches to evaluating such an effect can indeed be useful to the market.

However, a much more difficult question is the following: should the state standardize the methodology of organizing generative development itself?

In a market economy, the way production is organized is one source of competitive advantage.

One IT company may build development around several specialized AI agents. Another will create its own architecture. A third will use open models. A fourth will find that for a certain class of tasks the массов adoption of agents is economically unjustified.

The result of this competition should show which models are truly more effective.

Therefore, when forming future standards, it is crucial to separate two things.

The first is requirements for the outcome: security, quality, reliability, comparable performance indicators, and correctness of assessing the economic effect.

The second is the way of achieving that outcome: the solution architecture, the model used, the development environment, and the organization of a company’s internal processes.

In the first case, standardization can help the market.

In the second, it can, on the contrary, limit technological competition.

If a national standard begins to закреплять a certain architecture or a specific model for organizing generative development, there is a risk that the market will move not toward the most effective solution, but toward the solution that better matches a заранее established methodology.

Who should create the rules of the new market?

There is also another fundamental question.

The инициатор of creating the Council was Sber—at the same time one of the country’s largest technology players, actively developing its own models, tools, and AI infrastructure.

Sber proposes transferring to the Council participants the накопленный experience of transforming the development processes of tens of thousands of its engineers as a starting base.

This experience is undoubtedly of great value.

But the experience of the largest corporation is not necessarily a universal model for the entire economy.

A small IT company, an industrial enterprise, a regional developer, and a large technology corporation differ in resources, computing infrastructure, team scale, and project economics.

Therefore, the corporate practice of any market participant—Sber, Yandex, VK, or a small technology company—can be a source of data and expertise, but should not automatically become the model for a national standard.

It is especially important to ensure real, not nominal, participation of various categories of developers and customers.

Wording about the participation of the scientific community, customers, and developers has already been voiced. However, for the market it is far more important to understand how exactly the Council’s composition will be formed, how decisions will be made, how well small and medium-sized technology companies are represented, and how alternative technological approaches will be taken into account.

A standard must not become a hidden barrier

There is another risk that must be considered in advance.

A national standard in itself does not necessarily mean a ban on using other technologies.

However, the significance of the standard changes sharply if compliance with it later becomes a condition for receiving government support, participating in procurement, working with state-owned companies, or obtaining other advantages.

In this case, even a formally voluntary methodology may turn into an economic barrier.

That is precisely why the rules must be designed before such consequences appear, rather than fixing the market after they have arisen.

Supporting Russian technologies does not mean protecting them from competition

This discussion goes far beyond generative development.

Today, the state is simultaneously solving two different tasks: developing the domestic technology industry and increasing the efficiency of the Russian economy through the adoption of new technologies.

These tasks are related, but they are not identical.

A developer may indeed require state support to create a competitive Russian product, conduct research, and overcome technological barriers.

But the task of an enterprise buying technology is different—to increase productivity, reduce costs, improve product quality, and remain competitive.

If these two goals are mixed, there is a danger of evaluating the success of technology policy by the number of domestic solutions implemented rather than by the economic outcome achieved.

In our view, a strong Russian technology industry should be formed not by eliminating competition, but by the ability to win this competition.

The task of the state is not to choose a technological winner for the market, but to create conditions under which a Russian winner can emerge.

The Association of Innovative Solutions and Artificial Intelligence “Regions of the 21st Century” считает it necessary, when developing methodologies and future national standards in the AI sphere, to proceed from several principles:

technological and vendor neutrality; open participation of various categories of developers and customers; separation of requirements for the outcome and requirements for the means of achieving it; independent assessment of economic efficiency; inadmissibility of turning standards into an instrument for restricting competition.

And before creating a new mandatory or de facto mandatory regulatory mechanism, it is necessary to answer a simple question:

What specific market problem are we solving, and why is the market unable to solve it on its own?

If we are talking about security, compatibility, data protection, or comparability of performance indicators, the need for common rules may be justified.

If, however, we are talking about how a private company can more effectively organize its own development process, the state should approach unification with extreme caution.

Artificial intelligence is developing too quickly to administratively fix a single correct model for its use today.

Russia needs its own strong technologies.

But even more, it needs an environment in which these technologies become strong through efficiency and competition, rather than through заранее established rules of selection.