Overview
- Late‑June surveys found 44.7% of organizations run AI-generated code in production and about 35% use AI but keep its outputs out of shipping because they lack confidence in visibility and controls.
- Teams rely on AI mostly for low-risk, repetitive work such as documentation, unit tests, simple functions, and code review where errors are easier to catch.
- Respondents say the riskiest AI-introduced changes are the hardest to spot week to week, especially security flaws, dependency shifts, and performance regressions.
- Companies are buying layered safeguards—code quality analysis, automated review, software composition analysis, and security testing—and more than four in five have adjusted development or release processes for AI code.
- Developers report more time spent validating AI outputs and worry about long-term maintainability and lost learning for junior engineers, which could push firms to invest in governance, traceability tools, and new training approaches in the year ahead.