AI Agents Now Writing Most Code on Some Teams
The software engineering workflow is rapidly changing, with AI agents taking over coding tasks. At Anthropic, around 60% of engineers reportedly use Claude for over half their work. Some frontier teams even see the majority of their pull requests generated by AI agents, signaling a major shift in the role of human engineers toward system design and review.
The adoption of AI coding assistants is becoming nearly universal, with some studies showing 80-85% of developers now use them regularly. Across the industry, 90% of Fortune 100 companies are utilizing AI coding tools, and developers report saving an average of 3.6 hours per week. This trend is moving beyond simple code completion to fully autonomous agents. Cognition Labs, a startup funded by Peter Thiel's Founders Fund, introduced "Devin," which it billed as the world's first AI software engineer capable of handling entire development projects from coding to debugging and deployment. Devin's launch created significant buzz with demos showing it completing freelance jobs on Upwork and handling complex tasks. However, this was met with skepticism, as critics pointed out that some video demonstrations were misleading and that the AI often fixed bugs it had created itself rather than solving the core engineering problem presented. The rise of these powerful agents is directly impacting hiring at major tech firms. Data shows a 20% drop in employment for junior developers (ages 22-25), while job postings for machine learning engineers surged by 40% in 2025. Executives at Meta and Google have noted that AI is beginning to handle the work of entry- and mid-level engineers. Consequently, the most in-demand skills for software engineers are shifting. Proficiency in foundational languages like Python and Java is now table stakes; Big Tech recruiters are prioritizing expertise in AI/ML frameworks like TensorFlow and PyTorch, cloud infrastructure such as AWS and Kubernetes, and prompt engineering. The role of the human engineer is evolving from a line-by-line coder to an architect and system designer. The future of the job involves orchestrating suites of specialized AI agents, focusing on high-level problem-solving, and validating the output of AI systems. This new workflow is already showing massive productivity gains in specialized areas. At Nubank, for instance, using Devin for large-scale code migrations resulted in a 12x improvement in engineering hours saved and slashed project costs by a factor of 20. Despite the rapid integration, trust remains a significant hurdle. Only about a third of developers report fully trusting the code generated by AI assistants. Studies have found that AI-generated code can contain 1.7 times more defects if it is not overseen and reviewed by an experienced human engineer.
