The shift
53%
now spend more time reviewing than creating code
Up from 18% before adopting AI
Last year, our first State of Code Developer Survey found that the surge in AI-generated code had not yet delivered the productivity gains teams expected. It created a verification bottleneck instead, and the largest group of organizations still reviewed AI code the same way as code written by people.
This year, we asked over 1,000 developers and engineering managers what happened next. The short answer: agents made AI coding better, and code became abundantly faster than teams could build confidence in it.
This page highlights the key findings of the report. Download it for the full picture.
Download the full reportThe shift
53%
now spend more time reviewing than creating code
Up from 18% before adopting AI
The bottleneck
76%
generate AI code faster than they can verify it
The gap
39%
always check their AI-generated code before committing it
Down from 48% last year
The fix
77%
say strong verification tools make them trust AI code more
Pick your industry, company size or region, and key findings highlight the experience of your peers.
Your peer groupAll respondents1,026 respondents
Preview only · The full report is in the PDF Download the full report
A year ago, respondents told us AI generated or significantly assisted 42% of the code they committed. They expected that to reach 55% within a year and 65% within two. But it turns out they got to 65% in just one year.
Average, respondents who write code
What percentage of the code you personally committed or contributed was (or will be) generated or significantly assisted by AI tools / agents? (average, respondents who write code)
SonarQube users see better outcomes from AI coding.
Agents now write most of the code, but engineers need the confidence to ship it. This year's data shows SonarQube users are more likely to have that confidence, and more likely to say that AI has had a very positive impact on security, quality, and productivity. They are also more likely to have the practices this report keeps returning to: well-defined guidelines, systematic context for agents, verification in CI and measurement.
+81%
Code security
SonarQube users are 81% more likely to say AI-generated code has had a very positive impact on code security.
+65%
Vulnerability rates
SonarQube users are 65% more likely to say AI-generated code has had a very positive impact on vulnerability rates.
+42%
Defect rates
SonarQube users are 42% more likely to say AI-generated code has had a very positive impact on defect rates.
+46%
Technical debt
SonarQube users are 46% more likely to say AI-generated code has had a very positive impact on technical debt.
+66%
Code quality
SonarQube users are 66% more likely to say AI-generated code has had a very positive impact on code quality.
+39%
Productivity
SonarQube users are 39% more likely to say AI-generated code has had a very positive impact on their productivity.
“Very positive impact” only. Base: 319 current SonarQube users and 707 non-users (past and never users), all respondents.
7M+ developers and 75% of the Fortune 100 rely on SonarQube for AI code verification and governance.
AI now writes or assists with most code, volumes keep climbing, and a third of respondents expect to run software factories within a year. Yet generation is outpacing verification, manual checks are slipping, and returns are hard to prove. The teams with the strongest outcomes build confidence in their AI-generated code from the start.
We organize what they do around four terms:
In this formula, each of the elements of code confidence are multiplied, meaning, if any one of them is zero, then code confidence is zero.
23%
have well-defined AI guidelines or automated checks
54%
feed agents context through MCP or retrieval integrations
61%
of AI-generated code gets automated review, on average
36%
track AI's productivity impact and have insight into it
Code confidence holds when agents stay on track, issues are found before hackers find them, and reviews turn into fixes. The full report shows what the teams with the strongest outcomes do differently, and how Sonar helps make your agentic software factory better.
Missed our previous report? Check out last year’s findings here.