LLMs are the missing link to shipping side projects (valatka.dev)

A developer tested LLMs (Claude Sonnet 3.5 and 4 via GitHub’s Copilot VSCode extension) to speed through tedious front-end work for two side projects. The first experiment—asking the model to “surprise me” with a landing page—produced quick but generic results because the author lacked a clear vision and couldn’t give useful feedback. The second experiment, refactoring a personal blog into a cleaner landing/about layout, succeeded in minutes after iterative prompting and a light code review. The more capable Sonnet model helped, but the decisive factor was having a well-defined end state that guided the assistant’s edits. For the AI/ML community this reinforces a practical lesson: LLMs are highly effective at executing well-scoped implementation tasks (HTML/CSS, UI tweaks, forms, layout comparators) and can materially reduce procrastination and development time for side projects. However, they’re not a substitute for product vision or design judgment—models excel when humans provide concrete goals, constraints, and iterative feedback. Tooling that integrates LLMs into the editor (Copilot, etc.) is already a force multiplier for routine coding, but teams should treat them as powerful copilots, not autonomous creators.