On September 8, 2026, Yandex introduced Alisa AI for Business — a corporate AI assistant that executes multi-step work tasks using agentic scenarios and connects to external corporate systems through a plugin catalog.
What Happened
The assistant received a 'Tasks' mode, in which it operates using agentic scenarios: the system automatically breaks down a request into individual steps and sequentially executes chains of actions. Within the Alisa AI ecosystem, it accesses Yandex 360 services — Mail, Disk, Calendar, Tracker, and Wiki — and through the 'Plugins' section, it connects to external corporate systems: Bitrix24, SPARK, amoCRM, and 1C. The assistant is trained on the company's knowledge base — guides, manuals, templates, price lists, and brand books — so it operates taking into account the specifics of a particular employer. Access rights are configured by administrators, and Alisa's personal and work spaces are separated. The service can be used on the alice.yandex.ru website, in the Alisa AI mobile app, and through a widget in Yandex Mail. The product is available to all Yandex 360 clients — more than 185,000 organizations — and new users are provided with a three-month free period.
Context
Alisa is Yandex's mass-market assistant, and Yandex 360 is a set of corporate services (Mail, Disk, Calendar, Tracker, Wiki) used by more than 185,000 organizations; it is precisely through this ecosystem that the company is launching a new B2B-class product — an agentic assistant integrated into office processes. Agentic AI is a system that automatically breaks down a task into steps and executes them, unlike a classic chat interface that responds with text. In the office automation market, Copilot and Genspark-like solutions are already competing, and the 'ready-made agent without prompting' approach — when the user does not write special commands — is, according to experts, likely to become a reference for other office AI assistants.
Why This Matters for the Industry
Yandex is moving its mass-market assistant into the class of agentic systems that execute chains of actions and integrate with corporate software — this is a direct entry into the enterprise automation market, where Copilot and Genspark-like solutions are currently competing. The 'ready-made solution without prompting' model plus three months free for more than 185,000 clients turns the Yandex 360 ecosystem into a channel for mass distribution and the largest experiment on the readiness of the Russian market to pay for agentic office automation, to which competitors will have to respond within the trial window. The plugin catalog for 1C, Bitrix24, amoCRM, and SPARK sets the integration framework with corporate systems common in the runet, and in technical terms, this is scenario-based orchestration of actions on top of existing APIs, meaning a ready-made foundation for office automations without building an agent from scratch. As real usage accumulates, Yandex can be expected to publish task success metrics and plugin API documentation, which will make the product verifiable for engineers.
Why This Matters for Users
Companies can test agentic automation on real processes for three months free without writing prompts: the assistant automatically breaks down a meeting into tasks, searches for recurring issues in requests, analyzes tables, compiles reports on invoices, and works with Mail, Calendar, Tracker, and Wiki. For employees, it is important that Alisa's personal and work spaces are separated, and administrators control access rights — this is a standard condition for enterprise systems without which an AI tool is not implemented. For technical specialists, it is useful to study the approach to plugin integration of external systems and context separation as a reference for their own projects.
What Is Still Unknown / Limitations
Yandex does not disclose the base model, architecture, and training method, and public benchmarks and data on the success rate of multi-step chains are absent, so claims about agentic scenarios are currently confirmed only by product positioning, not by reproducible technical results. The training pipeline on the knowledge base — presumably a RAG-like approach to company documents — is not described: the quality of selection and the data update mechanism are unknown. There is no API for automation, only client interfaces, and reliability metrics, latency data, and prices after the free period have not been published.
Sources
Author
Look at AI, editorial team