Seven observations on real-world AI transformation in Russian business from two conferences

The author of the article, Zhemal Khamidun, CPO of AlpinaGPT and Head of AI at Alpina Digital, recaps observations from two online conferences held in 2025: 'AI-Transformatsiya' on April 3 and 'Tochka Sborki' on May 29. Across over twenty speakers and about 1500 registrations, several recurring themes emerged. First, vibe coding has become a corporate practice; examples include Ilya Berdysh of mymeet.ai, who without technical education built a macOS desktop app in two weeks using Claude, Conductor, and about 11,000 lines of code, and Denis Smetnev of Skyeng and uForce, who built five production tools without coding experience. Second, multi-agent systems are becoming the default, but there's a caveat about accuracy degradation with longer chains. Sergey Andriyanov of AlpinaGPT showed Evolver, a code review automation system, which moved from a single universal agent to three specialized agents, achieving three times faster reviews and 50-80 hours saved monthly. Third, data and processes matter more than models; a poll showed 90.3% of attendees believe pilots fail due to inflated expectations, data unpreparedness, and weak process integration. Artemy Malkov of Data Monsters emphasized rewriting back-office processes into Markdown and creating a Golden Dataset of rare cases. The author concludes with a formula: Tool × Training × Support = Result, and mentions external research by BCG and MTS Web Services.