# Copilot is making juniors mass-produce code they can't debug > Source: > Published: 2026-05-29 13:09:46+00:00 We have created a generation of developers who are able to write code at a pace that they cannot read. This shouldn't be a skill. By early 2024, over 1.8 million paid subscribers trusted GitHub Copilot to write their code. The tool is omnipresent. And no, I won't argue against AI assistants. I'll argue we've been turning a blind eye to one of their consequences, and it's going to sting. Copilot is optimized for one thing: output speed. Hit Tab, get code. Hit Tab again, get more code. It feels productive. It feels like flow. However, being able to follow the flow of code does not mean that you understood everything. A junior dev can scaffold an entire feature in an afternoon without understanding a single line of what was generated. The GitClear 2024 report revealed that there was more measurable code churn; that’s the code you write which is quickly revised or trashed. That’s not velocity. That’s waste. One uncomfortable truth that no one wants to address. When that AI-generated code breaks at 2am in production, Copilot isn't on call. You are. Stanford research has shown the tradeoff clearly: AI-assisted code ships faster but doesn't necessarily ship *more correctly*. Speed goes up. Bug rates don't go down. Sometimes they go up. I have seen developers spend minutes staring at a stack trace for something they wrote — I mean, they accepted. There’s a difference. If you can’t trace the logic in your head, you can’t make the fix under pressure. 🔥 The loudest debate in tech right now is "will AI replace developers?" Wrong question. The real risk is developers who can't function without autocomplete shipping production code to real users. That's not hypothetical. That's happening right now at companies of every size. Consider it in this manner: → A senior dev uses Copilot as a shortcut for boilerplate they already understand → A junior dev uses Copilot as a substitute for understanding the boilerplate at all → Same tool, completely different outcomes This doesn't imply that AI assistants are bad. It's an argument for being honest about the skill gap they can mask. I don't think the answer is banning Copilot. That ship has sailed. But teams need guardrails that nobody's building. The process of code review requires a shift in behavior. Reviewers should ask "can you walk me through this?" more frequently. If the author cannot explain the logical path, it's not a proper PR. Full stop. The onboarding process for new junior developers needs to be reimagined. They should dedicate their initial months to writing code “the hard way.” First, they need to establish the mental models. Then, introduce the power tools. You wouldn’t give a person a nail gun before teaching them how to use a hammer, right? **Interviews need to change.** If your hiring process tests whether someone can produce code quickly, congratulations — you're now testing whether they're good at using Copilot. That tells you almost nothing about debugging ability. 😬 AI programming assistants are amazing. They're also the first tools in software history that let you ship code you genuinely don't understand at a pace that *feels* responsible. This is a new combination that we haven’t developed cultural norms for. In the upcoming five years, what will set successful developers apart, is not their ability to write excessive amounts of code, but their ability to debug and understand the code produced by others. They'll be the ones who can debug what everyone else generated. And the scarcity commands high salaries. 💰 💰 **So here's my question: has your team changed anything about code review, onboarding, or mentorship since AI assistants became the default? Or are we all just pretending the Tab key is a teacher?**