AIが「弱い」エンジニアの害を軽減する
AIによるエンジニアリングのソフトウェアエンジニアリングの能力は「重い裾分布」であり、能力の低いエンジニアはプロジェクトに問題をもたらす「負の貢献」をすることが多い。
しかし、Claude Codeのような最先端の大規模言語モデル（LLM）は、この現象を変えつつある。
LLMは、単純なコードの欠陥（無限ループやデータ漏洩など）を自動で検知し、その品質の最低水準を大幅に引き上げた。
その結果、低スキルのエンジニアの作業成果は、かつてのような重大な問題を引き起こすことが減少し、そのアウトプットはより実用的なものになっている。
ただし、これはAIに依存するあまり、エンジニア自身が学習機会を失うという懸念も残る。
原文の冒頭を表示（英語・3段落のみ）
Like other kinds of puzzle-solving, software engineering ability is strongly heavy-tailed. The strongest engineers produce way more useful output than the average, and the weakest engineers often are actively net-negative: instead of moving projects along, they create problems that their colleagues have to spend time solving. That’s why many tech companies try to build a small, ludicrously well-paid team instead of a large team of more average engineers, and why so far this seems to be a winning strategy.
Being effective in a large tech company is often about managing this phenomenon: trying to arrange things so that the most competent people land on projects you want to succeed, and the least competent are shunted out of the way1. For instance, if you’re technical lead on a project, you more or less have to ensure2 that the most critical pieces are in the hands of people who won’t screw them up (whether by directly assigning the work, or by making sure someone can “sit on the shoulder” of the engineer who you’re worried about).
Claude Code changed this. Frontier LLMs don’t have the taste or the system familiarity of a strong engineer, but they have absolutely raised the floor for weak engineers. Instead of getting a pull request that could never possibly work or would cause immediate problems, the worst you’ll now see is a standard LLM pull request: wrong in some ways, baffling in others, but at least functional on the line-by-line level and not so obviously incorrect that someone with no knowledge of the codebase could point it out. That is a huge improvement!
※ 著作権に配慮し、引用は冒頭3段落までです。続きは元記事をご覧ください。