the participation of artificial intelligence agents in business processes has already become a fact / News of the economy of Krasnoyarsk and the Krasnoyarsk Territory / Newslab.Ru

While top managers of Russian companies are enthusiastically mastering no-code, technical specialists are increasingly looking more closely at the costs of mass implementation of AI agents.

At the Data Fusion forum, the speakers of the session “There will be no software: how agent-based AI is changing the architecture of corporate systems” were actually divided into techno-optimists and “old believers” IT specialists and discussed how much freedom can and should be given to AI agents in a corporate environment, and who will be responsible for their mistakes.

Head of Artificial Intelligence at T1 IT Holding Sergei Golitsynwho moderated the discussion, emphasized that AI agents have already “grown” into office software.

“In T1 there are processes in which multi-agent systems themselves formulate tasks, set subtasks for each other, develop solutions and verify them, essentially closing the entire product creation loop. The main question today is what tasks we are ready to delegate to such “silicon colleagues” who have already outgrown the level of a script with a neural network layer, and how ready developers and users are for these changes,” noted Sergei Golitsyn.

Senior Managing Director of the Moscow Exchange for retail business and the Finuslugi marketplace Igor Alutin emphasized that the penetration of AI radically changes the work with information: both the quality and cost of calculations directly depend on the quality and systematization of data. The evolution of AI agents in Moscow Exchange retail, according to him, began with FinGPT is a virtual financial advisor for private investors. Reliance on open sources quickly revealed the limits of quality: the answers were not always accurate, and the model had to be deliberately “simplified” by strictly defining work scenarios.

This experience convinced the team that a professional AI agent is impossible without relying on verified sources. Now the showcase of the “Finanuslugi” marketplace is being rebuilt for agent architecture, and Igor Alutin expects that over the next year up to half of user traffic will shift from classic search engines to GPT-like interfaces, including domestic ones.

“Purchasing financial products through an AI agent will become standard behavior for the client, but the need for human competencies when making investment decisions will not disappear. AI agents will take over the collection and primary summary of data, and humans will have to do verification for some time. As accuracy increases, people will move away from reading reports to checking and interpreting data, because the institutions of trust for investors are here to stay,” commented Igor Alutin.

As the meeting participants noted, the introduction of AI agents in companies almost always faces resistance inertia. Advisor to the General Director of R7‑Office Mikhail Alekseev admitted that “Old Believers” IT specialists will not disappear in the office environment, but their share will decline naturally – just as a cohort of people once disappeared, who used the Internet without a browser.

According to Mikhail Alekseeva, a change in user experience is already underway: young generation of IT specialists and office workers does not write macros in VBA, they initially focus on the natural language of instructions and instant results.

“In the future, 20–25% of users will access spreadsheets and other office tools through a conversational interface, immediately receiving insights, not just numbers—and this will become the norm for both novice professionals and board members. At the same time, AI does not provide sacred knowledge: its task is to relieve a person of routine. According to the company, 30% of the average office employee’s working time is routine operations that can be outsourced to AI assistants and agents. freeing up employees' time for reflection and creativity. Already now, one of the most popular functions is the intelligent semantic search of documents within individual workspace of the employee, in different channels of work communication: only this can free up up to 10% of working time and be directly converted into saving money,” concluded expert.

The voice of the traditional Russian IT school has become Alexey Lukatsky, business consultant on information security at Positive Technologies. According to him, in the current digital landscape, most companies have assistants, and there are still few truly autonomous agents who make decisions without human intervention, although their share will grow. He agezil against the demonization of security officers as “priests-Old Believers” who only prohibit everything.

“The real challenge for security is to embed technology, including AI agents, into a company’s circuitry so that it does not cause harm and fits within agreed upon levels of risk,” emphasiszero Alexey Lukatsky.

The paradox is thattill expert that AI has already penetrated deeply into critical infrastructure, including military applications. Without disclosing details Alexey Lukatsky gave an example of one organization that was faced with the task of hacking a combat torpedo with AI in order to show the attack vectors available to the enemy.

From a legal point of view, recallNile it, any AI function implemented in a financial organization or state corporation, automatically finds itself in the critical infrastructure circuit, even if it does not formally involve critical processes. In this case, the responsibility of the agent developer and security architect cannot be limited to convenience. WITHFirst, we need to put things in order in people’s rights and accesses, and only then build a secure “agent” architecture.

Director of Machine Learning Department MW.S. AI (included in MTS Web Services) Daniil Kireev described the risks of an uncritical passion for agents and no-code programming. Using specific examples, he showed how dangerous it is to leave development entirely to AI. This has a lot to do with the issue of AI hallucinations and “unknown unknowns.”

“Critical circuits require a multi-level architecture: separate censor models, extensible benchmarks for evaluating results, automatic testing and constant work on the safety and interpretability of agent behavior,” noted expert.

Summing up the session, the participants agreed on one thing: the participation of AI agents in business processes has already become a fact. The key skill of IT specialists in the coming years is not knowing “which buttons to press,” but the ability to formulate goals, limitations and quality criteria for artificial intelligence and take responsibility for the result of joint work between humans and AI. Given that even in large corporations the share of employees who regularly use new tools is still far from 100%, the ability to set tasks and manage agents can become the main competence of a manager in the era of collaboration between a person and a “silicon assistant.”