Developers Prefer Single-Agent AI Workflows, Keep Human Oversight, Survey Finds

A recent pulse survey conducted by Stack Overflow reveals evolving trends in the adoption and use of AI agents in software development. The survey highlights a significant increase in AI agent usage, which has almost doubled to 59% since 2025. However, this growth is accompanied by a strong preference for human oversight, as 63% of technologists report rarely or never allowing AI agents to operate completely autonomously. The findings indicate that the majority of developers, specifically 68%, favor single-agent workflows due to their predictable nature, rather than embracing more intricate multi-agent configurations. Among full-stack developers, GitHub Copilot stands out as the preferred single-agent tool, chosen by 65% of respondents. Other popular coding agents include Claude Code. Despite the growing integration of AI into daily work, developers continue to express concerns regarding the accuracy and security of AI-generated code. These two factors remain the primary barriers to broader adoption and trust in agentic systems. Interestingly, the perception of cost as a significant barrier has decreased, suggesting that the value proposition of these tools is becoming clearer to users and organizations. The survey also points to a trend where 60% of respondents actively prevent agents from making unapproved system changes, underscoring a cautious approach to AI integration. This suggests that while AI agents are enhancing productivity, the industry is settling into a model where human review and intervention remain critical components of the development lifecycle, ensuring quality and mitigating risks.