Alfa-Bank describes "vibe leadership": how AI took over routine work and sped up growth at the A-Token platform

Alfa-Bank executive Andrey Kalinin has described how the bank's in-house AI, AlfaGen, changed the way teams work on the A-Token platform. He says automating documentation removed part of the manual workload, sped up processes, and freed product leaders to focus on architecture, strategy, and scaling the business.

From routine to management

Kalinin ties this shift to what he calls "vibe leadership." The point isn't a trendy label but a reallocation of a manager's time: fewer hours go into mechanically assembling documents, chasing approvals, and rewriting the same boilerplate, and more go into the decisions that actually move the product forward. This matters especially in fintech, with its heavy regulatory and technical load, because documentation there isn't a by-product — it's part of the production pipeline.

At A-Token, the impact went beyond convenience. The team built a digital financial assets platform from scratch, then grew from 7 to more than 180 IT staff and, Kalinin claims, captured 50% of the Russian digital financial assets (DFA) market. At that scale, any routine task multiplied across dozens of people and processes quickly becomes an expensive drag on the business. And the more complex the system, the more every extra approval cycle costs.

What AlfaGen does

The main practical use case is generating and maintaining technical documentation in AsciiDoc. According to Kalinin, the in-house AI helped produce large volumes of documents, save millions of rubles, and cut time spent by roughly 20–25%. For product and engineering teams, this means some of the work can now move from a manual mode to a managed, semi-automated process — especially where documents live alongside code, releases, and internal policies.

• Drafting technical documentation in AsciiDoc
• Quickly updating existing specifications
• Standardizing document structure and language across teams
• Reducing manual workload for product managers, analysts, and tech leads
• Speeding up the rollout of new internal artifacts

"AI takes on the routine, and you focus on architecture, strategy, and

whatever brings in money."

That formula applies to more than just text. Once documentation stops being a bottleneck, approvals move faster, new hires onboard faster, and context passes more smoothly between product, engineering, and the business. In this setup, AI doesn't act as an autonomous manager; it's a tool that takes load off the most expensive specialists. As a result, quality control over decisions also speeds up. It also reduces the cost of communication errors.

Scale without overload

What makes the A-Token story interesting is that this isn't an experiment on a small team but a mature fintech business with strict requirements for quality and speed. When an organization grows from a handful of people to hundreds, knowledge management becomes a problem in its own right: rules have to be written down, changes communicated quickly, and decisions kept from getting lost in chats and calls. In that environment, AI pays off not only in hours saved but also in less chaos.

Kalinin's argument, though, goes beyond "let's have a neural network write our docs." He's talking about a change in the product leader's role itself. A strong manager used to be the person who personally pushed through texts, specs, and formalization; now their value is shifting toward systems thinking: where the platform is heading, which constraints are critical, where a new service is needed, and where to cut unnecessary complexity. For Russian fintech, it's also a sign that internal AI tools are maturing and being built into day-to-day operations where mistakes are costly.

What it means

The approach Alfa-Bank describes makes a simple point: the next corporate payoff from AI won't come from flashy demos but from taking routine work off the people who make key decisions. If this model takes hold, product leaders will write less by hand and spend more time managing architecture, priorities, and business growth. That's where banks and large platforms find a practical, rather than decorative, use case for deploying models.