AI and Software Engineering, the Conflict Within (handmadeoasis.com)

A developer reflects on months of heavy LLM use and argues that the rising clash between AI tools and software engineering is less about capability and more about what programmers value. They propose two broad categories: those who treat coding as a means to an end and readily welcome AI to produce results, and those who code for exploration, learning and craft—who feel robbed when an LLM does the creative work. The author finds LLMs helpful for repetitive tasks, boilerplate, or generating known solutions (a “compressed search engine”), but harmful when used to scaffold new systems or replace the exploratory work that helps them internalize a codebase and learn domain specifics. The piece matters to the AI/ML community because it highlights human, workflow and product-design implications beyond raw model performance: tools that automate should preserve opportunities for learning and explainability, or offer modes that support exploration vs automation. The author also notes a nascent third category—people who enjoy prompt-engineering, agent orchestration and “getting the LLM to output what they have in mind”—and worries some gains there may eventually be baked into core models. Practical takeaways: adopt LLMs selectively, design interfaces that expose reasoning and scaffolding, and recognize that sustainable adoption depends on aligning AI capabilities with the diverse motivations of developers.