Russian companies are struggling to move AI agent projects beyond the pilot stage due to poorly defined internal processes. According to a study by Comindware, businesses often fail to articulate clear operational rules, which are essential for AI agents to function effectively. Gartner forecasts that over 40% of agentic AI projects will be terminated by the end of 2027.
To overcome these hurdles, companies are shifting toward transforming static regulations into actionable instructions known as skills. This approach allows businesses to update operational rules via simple text rather than complex programming. In December 2025, Anthropic released the Agent Skills specification as an open standard, which has since been integrated into over 30 tools, including those from OpenAI, Microsoft, and Google.
Comindware has adopted this methodology within its own CRM system, where sales qualification criteria are stored as skills in a knowledge base. CEO Vitaly Shpak notes that this process forces companies to formalize their workflows, revealing gaps previously hidden by informal agreements. He predicts that corporate libraries of skills will become standard within two years.
Shpak emphasizes that as AI models become increasingly commoditized, competitive advantage will shift toward the quality of a company's digital instructions. The ability to precisely define internal operations will determine success, as the effectiveness of an AI agent is fundamentally limited by the clarity of the rules it is provided.
