Skip to content
back to the archive page
#AI

[2/2] AI in the SDLC: IT One and Skolkovo’s Study (Category AI)

#AI #Software #Engineering #Productivity #DevEx #Management #RnD #Leadership #Economy

Continuing my discussion and turning to Russia, I’d like to quote the report:

AI adoption in developers’ tasks in Russia is currently progressing at a pace comparable with the global market. Unlike globally, however, the problems and challenges of AI-driven changes to the development lifecycle are only beginning to enter the priorities of Russian technology leaders and team leads.

The report supports this with RUSSOFT’s 2024 study. In its software-company survey, 54.8% said they had AI expertise and experience and used it to build new custom systems or their own solutions, up from 46.6% in 2023.

The study’s authors estimate that around 62% of IT-team employees use AI, twice the 2024 level, and forecast roughly 98% penetration by 2028, or around 3.22 million users. I could not find the source of that forecast. They mention the RUSSOFT report above and a TAdviser labor-market report.

The authors then describe Russia’s AI-tool landscape and cite optimistic figures:

  • Developer assistants include Sber’s GigaCode, Yandex Code Assistant / SourceCraft and MTS Kodify; T-Bank’s Safeliner covers DevSecOps.
  • Public claims about these tools include:
    • GigaCode: up to 28% faster development, 50% less regression-testing time, 30% faster automated UI tests, 3 times faster pipeline configuration and 2 times faster DevOps onboarding.
    • T-Bank’s Safeliner: 5 times faster handling of security vulnerabilities.
    • SberTech’s Platform V Works: 40% less time to production in CI/CD, 50% less testing time, 25% more releases and 20% less effort.
    • Alfa-Bank’s testing agents: 30% fewer errors through broader coverage, up to 70% faster automated-test generation and deployment across 60+ teams.

The trends they identify are:

  • AI embedded in development platforms, including IDEs, CI/CD and DevSecOps.
  • Agent workflows for generating tests and code and conducting code reviews.
  • Changing expectations of efficiency and further movement toward T-shaped developers.

The challenges are interesting too:

  • Measuring impact: fewer than 25% of companies do so. Those that do focus on time to market, lead time and production defects.
  • Regulation and the quality risks of nondeterministic LLM behavior: information security, Russia’s 152-ФЗ law, model quality, usage controls and hallucinations.

The authors propose this adoption plan:

  • Choose 1–2 narrow use cases, such as unit tests or code review.
  • Record baseline metrics: time to market, lead time and defects.
  • Set an information-security policy covering permitted uses and on-premises versus cloud LLMs for code.
  • Integrate AI into the development platform and CI/CD rather than bolting on improvised tools.
  • Introduce quality gates for AI code: style checks, static analysis and mandatory review.
  • Build a prompt library and train the team.
  • Compare Russian and international models on your own tasks.
  • Scale after 2–3 successful pilots with demonstrated ROI.

#Software #Engineering #Productivity #DevEx #AI #Management #RnD #Leadership #Economy