Software development as a philosophical act

Marco Sbragi's essay on Dev.to argues that software development should be treated as a philosophical practice, focusing on structural reasoning rather than syntax [Dev.to]. He recounts teaching Visual Basic in the early 2000s, where students learned to reason about constraints and architecture on their own. Sbragi applied this approach to a recent project, GemmaLink, built on Google's lightweight Gemma models, using an LLM only for boilerplate syntax. He contrasts this with LLM-generated code, citing GitHub's 2025 Copilot report, which found that 45% of AI-generated snippets needed manual correction [GitHub Blog].

The importance of structural reasoning is evident in the fact that LLMs generate statistically probable token streams that lack contextual judgment. By prioritizing high-level design, developers can integrate AI tools as accelerators rather than replacements, delivering products faster while maintaining long-term maintainability. For instance, Sbragi's use of Go for GemmaLink, despite limited prior experience, allowed him to meet the zero-dependency requirement and retain full control over data layout, testing, and debugging.

A balanced approach, where reasoning defines the problem space and LLMs fill in the syntactic gaps, can deliver both speed and depth. Over-emphasizing abstract thinking risks slowing delivery cycles without clear ROI, while neglecting philosophical aspects can lead to brittle AI code and technical debt. By combining architectural thinking with AI assistance, companies can gain a competitive edge and preserve talent, as junior engineers learn to frame problems before reaching for syntax.