Vibe coding and agentic engineering are getting closer than I'd like (simonwillison.net)

In a recent podcast episode, Simon Willison discussed the convergence of "vibe coding" and "agentic engineering," which has raised concerns regarding the accountability and quality of AI-assisted coding practices. Vibe coding allows users, often non-programmers, to create software without needing to understand the underlying code, relying on AI tools to deliver functional outputs. In contrast, agentic engineering emphasizes professional standards, including security and maintainability, leveraging AI tools to enhance human expertise. Willison expressed discomfort over blurring these two distinct approaches, especially as AI's reliability grows, leading him to not review generated code thoroughly before using it in production. This shift poses significant implications for the AI/ML community and software development at large. As the capabilities of AI coding tools improve, the entire software development lifecycle is evolving; it can now produce thousands of lines of code in a fraction of the time previously required. However, Willison highlighted the risk of diminishing quality assurance and the reliance on the "black box" nature of AI-generated code. He argued for a more cautious approach, suggesting that users should prioritize solutions that have been rigorously tested by others, as demonstrated by his preference for trusted software products over hastily produced ones. This evolution challenges traditional development processes and calls for new evaluation standards in software reliability and performance.