Learning Programming in an Age of LLMs

Mark Seemann responds to a reader who used LLMs to assemble a large TypeScript/JavaScript system only to find it beyond their ability to maintain or deeply understand. Seemann discloses he leans against AI, worries about programmer job displacement and the broader economic risk of mass knowledge-worker unemployment, and is skeptical that new technology-driven jobs will necessarily absorb those displaced. He suggests practical alternatives (trades that rely on manual skills) and emphasizes his continued personal study of fundamentals more out of curiosity than career calculus, arguing programming historically required understanding adjacent abstraction layers to troubleshoot effectively.

On learning and practice he advocates stepping back to systematically learn fundamentals when needed, while acknowledging that deep competence took him decades and that human learning speed, not access to materials, is the bottleneck. He learned through examples, documentation and books, and would approach new languages today by reading existing code and filling gaps. He uses LLMs selectively: for verifiable, falsifiable coding tasks and short code improvements, not for open-ended guidance, because models routinely hallucinate. His core warning: LLMs can accelerate building but also enable creating systems that exceed a builder’s understanding, producing fragile, hard-to-maintain software.