AI and Home-Cooked Software (mrkaran.dev)

AI is turning non-developers into builders: domain experts with no coding background are using LLMs (examples include Anthropic’s teams using Claude) to create “home-cooked software”—small, highly specific tools and automations built in hours or days instead of months. These personal apps—anything from a custom CSV reformatter or a tailored export format to a single-page calculator or a prom2grafana script that converts Prometheus metrics into Grafana dashboards—solve immediate, niche problems without the overhead of traditional productization. The result is a new top layer of software: messy, one-off, and hyper-personalized, but enormously empowering because the barrier from idea to working prototype has collapsed. That accessibility changes the economics of software creation but brings important technical caveats. AI can generate useful first drafts quickly, but moving to production remains hard: handling edge cases, ensuring security, and debugging take disproportionate time. Generated code can contain real vulnerabilities (SQL injection, hardcoded secrets, race conditions), hallucinated APIs, or brittle, inconsistent styles that create long-term maintenance debt. Prompt engineering and verification are time-consuming, and organizations will face governance challenges as incompatible personal tools proliferate. Still, for non-critical, personal workflows the tradeoff is often acceptable—more creativity and autonomy at the cost of fragmented, repairable technical debt.