The best line of code is the one you never write. That idea sounds like an excuse for laziness until you have spent a few years maintaining someone else's enthusiasm. Writing software feels like construction, but it behaves more like gardening. Left alone, a garden grows whether you want it to or not. Code does the same thing. The real craft is knowing when to stop planting.
Your Code Is a Liability
Every line you commit creates a set of ongoing obligations. You will read it again during a late-night incident. You will test it after your framework releases a minor version bump that changes string handling. You will debug it when logs from production make no sense. You will explain it to a teammate who joined last week, or to yourself twelve months from now when the context has evaporated.
This is not an argument for obscurity. It is geometry. Bugs need space to hide. The smaller your surface area, the fewer places failure can burrow in. A function with eighty lines and six nested conditions is not just harder to read; it is statistically more likely to surprise you. Restraint is not the absence of effort. It is the recognition that unwritten code has exactly zero defects.
When Cleverness Becomes a Tax
Consider the task of calculating an order total with a few business rules: apply a discount, check for taxable items, skip anything marked as removed. One developer writes a single expression. It streams the list through a complex filter chain, invokes a helper library, folds the result with a curried reducer, and returns the sum. It is compact. It might even be elegant in a academic sense. But to read it, you must understand the helper library's implicit casting, the order of operations inside the stream, and the business logic all at the same moment. You cannot set a breakpoint in the middle. You cannot drop a log statement without breaking the chain. The code is short on the page and long in the mind.
Another developer writes a basic loop. She declares a running total, iterates through the items, and uses a plain if statement to decide whether tax applies. The block is taller vertically, but the intent is obvious. You can read it from top to bottom without holding five abstractions in your head. You can step through it in a debugger. You can add logging on line four without refactoring the whole expression.
Clever code looks smart in a pull request for about ten minutes. Simple code looks boring, and boring is exactly what you want when you are troubleshooting at two in the morning. Your goal is clarity, not a demonstration of intelligence.
Systems Need Structure, Not Heroes
This principle scales up to architecture. A clever system might rely on handwritten consensus logic, bespoke orchestration scripts, and undocumented caching shortcuts that only one engineer truly understands. That system does not run on its own; it runs on the constant brilliance of whoever is holding it together. When that person takes a vacation or a new job, the system starts to wobble.
Well-designed systems rely on structure and constraints instead. They use database schemas that reject bad data, API contracts that define boundaries, type systems that catch category errors before deployment, and module separations that make the intended path obvious. They do not require heroics to stay stable. They are designed to survive contact with tired humans, which is the only kind of human who ever operates software in production.
The AI Amplification Problem
Artificial intelligence coding assistants make this lesson urgent. These tools generate text quickly. Present them with a simple problem and they will often return a large, complex solution that imports utilities you already have in-house wrappers for, handles edge cases that do not exist in your domain, and uses idioms from a framework version you migrated away from two years ago. The AI looks at the immediate task. You must look at the whole system.
長期的なコストを管理せずにすべての提案を受け入れてしまうと、コード生成はインフレを引き起こします。リポジトリは、コンパイルも通り、テストもパスする、一見もっともらしいコードで膨れ上がりますが、誰もそれを真に理解していません。危険なのは、目立つ構文エラーではありません。それらはレビューで発見されます。真の危険は、コードベースが徐々に肥大化していくことです。個々のファイルは単体で見れば妥当に見えますが、全体として人間の脳に収まりきらなくなります。エンジニアリングの速度は、こうして失われていくのです。衝突によってではなく、誰もが完全に把握していないものに触れるのを恐れ、削除することを躊躇するものが静かに蓄積していくことによって。
削除は設計スキルである
優れたエンジニアは、誰よりも速くタイピングすることで自分を証明するわけではありません。彼らはシンプルさを選び、蓄積よりも削除を選ぶことで勝利します。コードを削除するには、そのコードを理解する必要があります。データフローを追跡し、その機能に隠れた呼び出し元がないことを確認し、ビジネス要件がすでに変わっていることを検証しなければなりません。削除は追加よりも困難です。なぜなら、削除には確信が求められるからです。
チームはしばしば、リリースされた機能や、膨大なプルリクエストを投げるマルチプライヤーを称賛します。一方で、4,000行もの不要なロジックを削除し、システムをより高速で理解しやすいものにしたエンジニアを称賛するチームはほとんどありません。しかし、その「マイナスの行数」こそが、組織の未来に対するより大きな貢献となることが多いのです。
高価な部分
今やコードは安価です。誰でも数秒で何ページものコードを生成できます。高価なリソースは「明快さ」です。システムを理解可能な状態に保つには、時間、判断力、そして自制心が必要です。真のエンジニアリングは、下書きではなく、編集の過程で行われるのです。
書く量を減らし、削除を増やす。シンプルに設計する。