State of Code Developer Survey report by SonarSource (AI column)
Interesting. 57-page report producer SonarQubeMeans for verification of code quality. The report came out. 8 January 2026 It was based on a survey from October. 2025 years. To sum up in one sentence, the basic idea is this: AI didn’t take the workload off development – it just moved the bottleneck from writing to code verification.
The methodology of the study looked like a quantitative online survey 1149 Respondents around the world. The sample includes adults in tech roles who write code or manage developers and have used AI in their work over the past year. The questions were not just "do you use AI," but how exactly it changes engineering work: Where AI really helps Where there is a gap between acceptance (adoption) efficient (effectiveness) Do developers trust the AI code? How security and technical debt are changing What happens to junior/senior engineers and is there a difference in their attitude to AI?
- What about using AI agents
- What's going on with the management of this whole tool zoo?
The results came out like this. AI is no longer an experiment, but a routine. 72The percentage of those who have tried AI coding tools use them every day. 42% of code in commits is already AI-generated or AI-assisted The value of AI is unevenly distributed: AI performs best in documentation, explaining existing code and generating tests; noticeably worse in refactoring and modifying existing code. (Recall this survey was before a big shift in model capabilities in the fall 2025 year) Teams juggling on average 4 AI tools, and 35Percentage of developers use some of them through personal accounts rather than through work-sanctioned access.
As a result, personal speed increased, but trust did not appear. 96% do not fully trust the functional correctness of the AI code. 95Percent spend time reviewing/testing/fixing AI output 38% say that revising the AI code is more difficult than a human code, 48% always check the AI-assisted code before committing. Unsurprisingly, the most important skill in AI-era was the review and validation of AI code for quality and security.
When it comes to sick places, 57% are afraid of leakage of sensitive data 47Percentage of new subtle security vulnerabilities. In terms of technical debt, the picture is also dual: 88Percent saw at least one negative effect of AI on debt, most often “the code looks correct but unreliable” (53%) and redundant/duplicate code (40%). But 93A.I. helps with documentation, tests/debugging and part of refactoring work. Junior developers are more likely to rely on AI and are more likely to say that checking such code requires more effort than senior developers.
There is also a difference between organizations of different sizes: Small and medium-sized companies (SMB) They're taking more advantage of faster delivery and faster time-to-market improvements, and corporations are more likely to see improvements in code quality and support -- seemingly through tighter governance and AI risk control.
The conclusions of the authors of the report are simple – development processes need to understand the origin of code, strengthen code revision, add quality gates, static analysis, and build normal governance around AI tools. Otherwise, it is not the delivery, but the production of unverified code that accelerates. And of course, with all this can help SonarQube from the authors of the study:)
#Engineering #Software #DevOps #DevEx #Process #Management #Metrics