Skip to content
#AI

[2/2] AI research in SDLC from IT One and Skolkovo (AI column)

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

Continue. story About the study and moving to the Russian realities, I want to quote the report At the moment in Russia, the penetration of AI into the tasks of developers is at a pace comparable to the world. However, in the Russian market, in contrast to the world, the problems and challenges of changing the development life cycle under the influence of AI are only beginning to enter the priorities and focuses of technology leaders and team leaders. This quote is reinforced. russoft from 2024 A survey of software companies found that 54,8% of them indicated that they have expertise and experience in the field of artificial intelligence and use them in the development of new custom systems or their own solutions. (This number grew from 46,6percent 2023 year).

The authors of the study estimate that AI is used ~62% of IT team staff (into 2 higher 2024)eh 2028 It's going to be porn.98% (~3,22 million). I couldn't find the source of this prediction. (mentions the Russoft report cited above, and T-Adviser labour-market).

Next, the authors talk about the landscape of AI tools in Russia and share optimistic figures Developer Assistants: GigaCode (sber)Yandex Code Assistant / SourceCraft, MTS Kodify; DevSecOps: Safeliner (T-Bank) Public statements on the results of the use of these tools

  • GigaCode: up to +28% of the development speed; -50% of the time for regression; +30% speed of UI-autotests; 3× faster configuration of the conveyor; 2faster onboarding DevOps; -- 5faster processing of security vulnerabilities in Safeliner from T-Bank -- Platform V Works (SberTech.): −40% of the time before opening in CI/CD;50% of the time for testing; +25Percentage of releases; -20Percentage of labour input Alfa-Bank: AI-agents-testers30Percentage of errors due to increased coverage 70% faster generation of autotests; scale 60+

Among the trends, the authors highlight Embedding AI into development platforms: IDE, CI/CD, DevSecOps, etc. Emergence of agent scripts - autogeneration of tests / code, code revision Changing the standard of efficiency and further movement towards T-shaped developers

Interestingly, the challenges are topics.

  • Measurements of effects - less 25% of companies measure them, and those who measure based on TTM/Lead Time, defects on the product Regulatory and quality risks of non-deterministic work LLM - IB /152FL, LLM quality, control of use and hallucinations

The authors of the study propose the following AI implementation plan

  • Choose. 1–2 narrow scenarios: unit tests, code review, Record the basic metrics: TTM, Lead Time, defects
  • Accept the information security policy: what can / can not; on-prem / cloud LLM for code Embedded in the development platform/CI-CD, not artisanally sideways Enter quality-gate for AI-code% codestyle, statistical analysis, mandatory review Create a library of prompts and train the team. Compare Russian and foreign models on your tasks.
  • Scale up after. 2–3 Successful pilots with confirmed ROI.

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