A survey of 450 engineering managers at companies with 1,000+ employees found that AI agents — software systems that autonomously plan, write, and deploy code — are now responsible for approximately 30% of new code written at large technology companies. The figure, up from 8% a year ago, signals a fundamental shift in how enterprise software is built.
What AI Agents Are Actually Doing
The survey distinguishes between AI-assisted coding (where a developer uses Copilot or Cursor with human review at every step) and AI agents (where an AI system independently executes a multi-step task with minimal human oversight). The 30% figure refers to the latter category.
Common agent tasks include: writing and running test suites, fixing specific bug categories at scale, performing code migrations (e.g., updating deprecated API calls across large codebases), and generating boilerplate for new features based on specifications.
Implications for Software Teams
The survey data suggests headcount is not the primary impact — most companies report agent use is enabling teams to tackle work they previously couldn't prioritize, rather than replacing developers. 71% of respondents said agents had 'increased team output without reducing team size.'
The more significant change is in what developers do with their time. Engineers are increasingly reviewing and directing AI-generated code rather than writing it, shifting their work toward architecture, requirements definition, and quality oversight.
Risks and Concerns
Security engineers in the survey flagged AI-generated code as a growing source of vulnerabilities. 34% reported at least one security incident in the past 12 months that could be traced to AI-generated code that passed automated testing but contained subtle flaws.
Code review quality is also under pressure. As the volume of AI-generated code grows, human review capacity hasn't scaled proportionally — creating a gap where code is technically reviewed but not thoroughly examined.
The Takeaway
The 30% figure is striking and probably will look conservative in two years. AI agents writing and deploying production code are a real, current phenomenon at enterprise scale. The implications for software team structure, security practices, and developer skills are significant and only beginning to be worked out.