Why Copilot Is Making Programmers Worse at Programming
Claims that GitHub Copilot and similar AI code generators are making programmers worse spark sharp disagreement. Critics argue these tools erode core skills, encourage shallow understanding, and let weak developers pass as competent, especially in education and among juniors who may lean on them as a crutch. Supporters counter that, used thoughtfully, AI assistants act like advanced autocomplete or Stack Overflow, offloading boilerplate, accelerating learning of new stacks, and fitting into a long history of abstractions that changed how—rather than whether—developers need to understand low-level details.
Evidence and Speculation
- Many note the article offers no data, just plausibility arguments; some find the reasoning logical, others dismiss it as an “old man yells at cloud” rant.
- Several ask for proper studies; one link is shared suggesting worse code quality with Copilot, but others note conflicting studies even in that article.
- Consensus: current claims about long-term effects are mostly speculative.
Comparisons to Earlier Tools
- Repeated analogies to calculators, Stack Overflow, Google, IDE autocomplete, syntax highlighting, ORMs, high-level languages, and even writing vs oral culture.
- One camp says this is the same recurring panic and that the industry has always adapted.
- Others argue Copilot is different because it produces full solutions that can be used without understanding, unlike calculators that just do arithmetic.
Effects on Skills and Learning
- Concerns: erosion of core skills, less debugging practice, over-reliance on generated code, weaker fundamentals (already seen with SQL/ORMS, memory models, etc.).
- Several worry especially about students and juniors: easy cheating on basic algorithm assignments, or spending hours coercing an LLM instead of thinking through problems.
- Counterpoint: even if some low-level skills atrophy, that may be acceptable if higher-level productivity and new “AI collaboration” skills improve.
Developer Experiences with Copilot and LLMs
- Positive reports: major boosts in routine work—boilerplate, CRUD, tests, config, migrations, regex/glob snippets, new frameworks, code translation, explanation of unfamiliar code.
- Many use it as “super-autocomplete” or rubber duck: they still design solutions, then accept or edit suggestions.
- Negative reports: high error rates in anything non-trivial; debugging unfamiliar AI code can cost more time than writing it; tool can encourage bloated or repetitive code.
- Several emphasize that responsibility remains with the committer; the real risk is users blindly trusting large autogenerated changes.
Jobs, Standards, and Social Impact
- Some fear massive cost savings will reduce engineering headcount and wages, and allow weak developers or non-developers to flood the field with low-quality software.
- Others argue good engineering still requires judgment, empathy, and design skills that tools don’t replace.
- There is broad agreement that LLMs are here to stay; disagreement is about whether they mostly help, hurt, or just shift which skills matter.