Alexander Vedyakhin: “In the AI Alliance, we talk about «controlled autonomy»

The “Russian AI Center” booth at the World Artificial Intelligence Conference (WAIC) 2026 confirmed that businesses are moving from experimenting with artificial intelligence to fully delegating financially significant tasks to agents. The booth showcased two groundbreaking developments: the B2B platform GigaNetwork, where AI agents from different companies negotiate directly with one another, and the self-developing agent GigaAgent, whose evolution visitors observed in real time.

Commenting on the exhibition’s content in an interview with the Chinese media outlet GeekPark, Alexander Vedyakhin, First Deputy Chairman of Sberbank’s Management Board, noted that today’s agent is a digital executor.

AI is already conducting negotiations with suppliers on its own and finalizing some procurement deals. The key difference lies in the architecture of trust and the “tangibility of the decision.” If an agent handles a routine task—such as booking tickets—the cost of an error is low. But when it comes to issuing a loan, the agent acts as an economic agent by proxy, and in this case, the cost of an error increases many times over. For example, at the end of 2025, Sber issued its first fully autonomous loan to a corporate client—without any involvement from a bank employee, from application to decision.

Vedyakhin also noted that the future of AI lies in the convergence of agent systems with real-world infrastructure: energy and manufacturing. However, the main question on developers’ minds today is the level of permissible autonomy for machines. This issue is also being discussed

Alexander Vedyakhin, First Deputy Chairman of the Management Board of Sberbank:

“In the Alliance, we talk about ‘controlled autonomy.’ In complex production chains or healthcare, the cost of error is too high. Currently, models lack the ability to set their own goals, so the task here is not a race for AGI, but the creation of ‘practical, hands-on’ AI. We are restructuring the product development lifecycle (PDLC) around human intent, rather than around code. Only when we have accumulated reliability statistics on routine tasks will agents be ready to move into critical infrastructure.”

The AI Alliance is a key Russian organization that brings together the country’s leading technology, industrial, and financial companies for the joint development and implementation of artificial intelligence. Within the country, the Alliance focuses on specific practical steps. These include the creation and development of standards and codes of ethics for the application of AI in critical sectors. For example, a Code of Ethics for AI in Medicine and Healthcare has already been adopted in collaboration with the medical community.