T-Sync Conf and the AI4SDLC Study 2025 from T-Bank (Category Engineering)
Today, I'm going to spend the whole day at T-Sync, where there's no talk but more discussion and space. In fact, I'm going to have an area on the second floor where we can talk about AI4SDLC, which we launched last year with a community survey. The study itself is available on the website ai4sdlc.tbank.ru by clicking on the tab "Study" It consists of two parts:
1Section with meta-examination n 2023–2025where we took order 50 Some studies in recent years and looked at their results. It was especially interesting to look at the trends of serial studies that are conducted annually – they show how the attitude to AI in the development environment has changed, the penetration of AI in different scenarios, the assessment of its effects, and so on. Among such serial studies were studies from: DORA, GitHub, Stack Overflow, JetBrains, Atlassian, McKinsey, Jacob and Partners / Yandex, Express42, MIPT. There's a website. seriesYou can see the main trends of each series. There is also timeline With all the studies taken into account + you can study a brief sammari for each of them.
2Section with surveyWhich we did at the end of last year. The survey was extensive and I would like to thank all those who participated. According to the results, we started to fill out a survey of about 1k people and it is good that we made questions about the use and impact of AI at the very beginning, since not everyone reached the end of the survey:) Speaking about the profile of respondents, it turned out as follows:
- 50% developers, 17% technical managers, 7% of top managers, 6% qa engineers, 4% AI/ML engineers and so on
- 24% are employed by large companies (1k - 10k), 23% in very large (10k+ employees), 18per cent (100 - 500 staff), 16% in small (10-100)
- the main industries represented: finance - 25%, technology-- 20%, retail and e-com - 15%, telecom- 7%, education - 5% The main applications/services on which respondents work are distributed fairly evenly: internal / external for users. 45% vs 37% and b2c vs b2b - 34% vs 41%
As for the executive summary of the survey results, they are AI to generate/automate code is used "often/always" 58% for code review 24% for modernization of legacy; 18% (upon 42% "never"). Productivity has increased at 64% (18Percent "significantly")The ability to write code has improved. 37%.
- Code quality: improvement 32%, deterioration 14%, unchanged 42%. Trust in AI code is low: 49% don't trust, 11Percent trust (The rest are neutral.). The role of AI within the company has grown 54%; transparency of implementation plans is felt 50%, do not feel 21%. "System readiness": commit→prod for a noticeable part is measured in weeks/months; documentation as a source of truth works poorly (near 39–42% do not rely on it in critical situations).
These results respond well to the meta-study results AI coverage is approaching "almost everyone" (for example Dora to 2025 about 90%)In this case, the personal effect is usually stronger than the command. Bottlenecks move to integration, testing, review, releases, communications; research regularly surfaces the thesis that coding is only part of the engineer’s time (And that organizational barriers eat up a substantial portion of the week.). Quality and trust are the central issue of maturity: "almost right" answers create a new duty of vetting and debugging. Experimental data show that on complex tasks, the effect can be zero or negative; AI is a system amplifier, not a replacement for engineering.
- No guardrails. (tests, quality gates, small batch, review, observation) “Acceleration” easily becomes “acceleration of chaos.”
In general, if you are interested in this topic, then come to the stand and I will tell you more about this machine and answer additional questions.
#AI #RnD #Software #Engineering #Management #Leadership #Metrics #Processes #Productivity