AI Assistants Now Write Most Boilerplate

Boilerplate is the code nobody brags about. Config files, test scaffolding, the fortieth data class, the try-catch block around a network call. It is the part of the job that is necessary, repetitive, and completely unmemorable. Over the past three years, that work has quietly moved to the machine. Ask any engineer who has shipped something recently, and they will tell you the first draft of the dull parts now arrives written.

The shift has a short history. GitHub Copilot opened as a technical preview in June 2021 and became generally available a year later. By 2023 the Stack Overflow developer survey had added a whole section on AI tools. On an Alphabet earnings call in October 2024, Sundar Pichai said that more than a quarter of all new code at Google was generated by AI and then reviewed by engineers. Three years, from novelty to normal.

Why boilerplate first? Because it is the easiest thing in the world to predict. A model trained on public code has seen countless versions of the same pagination helper. The answer is not clever; it is average, and average is exactly what you want for a config file. The hard problems — the ones with a business rule hiding inside them — are still where humans earn their salary.

The day-to-day effect is not that engineers write less code. It is that they read more of it. A generated file looks finished before anyone has checked whether it is right. Teams that handle this well have added a small ritual: whoever opened the pull request has to walk through the generated parts and say which lines they actually understand. It sounds slow. It is faster than the alternative.

There is an open worry about people starting out. Boilerplate used to be how you learned the shape of a codebase: you typed the same route handler twenty times and eventually stopped needing to look up the syntax. If the machine types it, where does that muscle come from? Nobody has a confident answer yet. The teams thinking hardest about it are giving each new hire a few deliberately tool-free weeks.

The word itself is older than software. In the nineteenth century, boilerplate meant the steel plates used to build boilers; newspapers later borrowed it for ready-made text shipped to small papers as printing plates. The meaning has held for a hundred years: text you use without rewriting. It is fitting that the first job we handed to a language model was the one already named after mass production.

So how do you bring up this topic with your team without sounding like a sales deck? Talk about the task, not the tool. "The migration script was generated, I reviewed it line by line, and I rewrote the date handling" tells your lead three useful things. "AI wrote it" tells them nothing they can act on. Say what you delegated, say what you checked, and own the result either way.