Last spring, a senior engineer at a Series B SaaS company told me about a code review that unsettled her. The junior developer — two years out of a bootcamp, sharp, shipping more tickets than anyone on the team — had submitted a clean, well-structured pull request. It handled edge cases she wouldn’t have expected from someone at his level. When she asked him to walk her through a particular generic type abstraction he’d written, he paused. “Honestly,” he said, “Cursor wrote most of that. I’m not totally sure how the inference works.”
She approved the PR. The code was correct. But she told me she went back to her desk with a feeling she couldn’t shake — the sense that she’d reviewed a product without meeting its author.
That feeling is spreading. GitHub Copilot, Cursor, Claude Code, and a growing fleet of AI coding assistants have made developers measurably faster. A 2023 study by GitHub found that developers using Copilot completed tasks 55% faster than those who didn’t. Startups are thrilled. Engineering managers are thrilled. But a quieter conversation is happening in Slack channels and 1-on-1s, and it goes like this: we’re producing code faster than we’re producing engineers.
The Apprenticeship That Built an Industry
For decades, software engineering had an apprenticeship model that worked almost by accident. Junior developers were handed the work nobody else wanted — writing CRUD endpoints, fixing CSS layout bugs, adding input validation, wiring up test fixtures. It was tedious. It was sometimes demoralizing. And it was, without anyone intending it to be, the curriculum.
The tedium was the point. Writing your hundredth boilerplate API
