AI coding tools普及でもソフトウェア品質が低下し続ける理由
Original: If coding has been solved, why does software keep getting worse?
Why This Matters
AI tool adoption alone cannot address the organizational incentives that deprioritize software reliability and stability.
AI coding tools that promise 100% code generation are becoming widespread, yet software quality continues to decline. A tech blogger points to real-world bugs in banking apps, Slack, car infotainment systems, and LG appliances as evidence that the tools alone cannot fix organizational and incentive problems.
Despite widespread claims that AI will soon write all code, software quality is measurably worsening. The author cites several recent examples: a banking app requiring three FaceID attempts before 3D Secure loads; Slack on macOS stealing window focus mid-typing and sending a git pull command to a group chat; an LG refrigerator warranty form that silently fails at submission; and a car infotainment OS update that introduced random reboots, input lag, and unresponsive turn-signal audio — a safety concern while driving.
The author acknowledges that nostalgia for 'stable' eras like macOS Snow Leopard is partly selective memory, and that complexity has grown with each new abstraction layer and framework. However, the core argument is structural: software teams may have access to the latest LLMs with generous token budgets, yet quality still declines because organizational KPIs reward new features over stability. As the author puts it, no PM at a large company can present a quarter focused exclusively on bug fixes. Until incentive structures change, the 'great software quality decay' will continue regardless of what AI tools are available.