Only 40% of solutions on the global market are full-fledged AI agents

Consulting firm Yakov & Partners analyzed 30 AI products and 65 public implementation cases globally and in Russia. The study 'AI Agents: Where Hype, Where Reality' showed that only 7 out of 18 international solutions (39%) meet at least five of six agent criteria. Among 12 publicly available Russian products, none were ready-made AI agents: four leading solutions are platforms for building agents, while the remaining eight are specialized assistants and chatbots. In corporate implementations, only 10 out of 43 international cases (23%) reached maturity levels L4–L5, capable of end-to-end process execution with minimal human involvement. On the Russian market, 4 out of 22 public cases reached L4, all developed internally by large companies. The study proposes six operational criteria for agentness: agent cycle, planning, tool use, self-correction, memory, and autonomous task completion. Products that passed the full-fledged agent threshold include Claude Code, Devin, Codex by OpenAI, Operator by OpenAI, Cursor, Harvey, and Manus. The report notes a gap between widespread generative AI adoption and financial impact: over 70% of Russian companies use generative AI, but only 9% report an effect exceeding 5% of EBITDA. Experts attribute this to a 'cascade of attenuation' where local improvements are hindered by approvals, integration, and legacy systems. The four most mature Russian publicly available solutions are Yandex AI Studio, GigaChat Enterprise by Sber, Agent Platform by Just AI, and MWS AI Agents Platform by MTS, all platforms for assembling agents. The study emphasizes that successful deployment requires correct solution type identification, appropriate pilot scale, and economic evaluation. It also highlights upcoming Russian legislation on AI, effective from September 1, 2026, which introduces sovereign and national model statuses, urging companies to plan architecture now.