{
  "id": 13270244,
  "title": "The Epistemology of Quality",
  "url": "https://urgent.news/2026/10/10/the-epistemology-of-quality",
  "topic": "tech",
  "section": "Tech",
  "published": "2026-10-10T00:28:26.000Z",
  "source": {
    "name": "Dev.to",
    "slug": "dev-to",
    "url": "https://dev.to/cheetah100/the-epistemology-of-quality-4cmi"
  },
  "original_language": "en",
  "account": "Should we expect human software developers to review code written by AI? For many, the answer is clearly yes: how else can we maintain accountability? However, there are concerns about the effectiveness of this approach. The benefits of traditional code review practices have diminished in modern software development. The question of how we know code is correct becomes more complex when dealing with real systems that have users, evolving requirements, dependencies, concurrency, and unanticipated failure modes. We cannot prove code correctness, but we can seek evidence. This article explores how code review evolved, the types of evidence each stage provided, and its limitations.\n\nIn-person code review has changed over time. In the past, code was printed and marked up with red pen during meetings, with developers discussing it. However, these reviews focused on syntax checking and coding standards rather than teaching. The benefit was the opportunity to cross-pollinate ideas and have developers contribute better implementations. Nevertheless, unit tests and discussions about the purpose of the software were lacking. Later projects shifted towards developers being responsible for quality, including ensuring code had no defects. Developers introduced unit tests, automated coverage analysis, and compulsory review by a second developer before committing. Reviewers sat next to developers, and pair programming became common. Pair programming involved discussing and reviewing code together before committing. This approach provided stronger evidence, as it combined various types of evidence, such as watching the feature run, passing tests, measuring coverage, and having a second pair of eyes read the code. Reading code was the weakest signal for finding defects, but having someone who understood the architecture could provide valuable judgment about structure.\n\nOpen source created challenges for source control, as developers were no longer co-located. Git encouraged separate branches for review, and the pull request became the place for review. Tools automated certain gates, such as running unit tests, maintaining test coverage, and finding common defects. The pull request model removed the side-by-side discussion of implementations, which was crucial for educating developers about quality. It also became too easy to approve a PR without thoroughly checking the code. The evidence for this shift came primarily from CI/CD, with build, unit tests, and static analysis happening on every commit. In an age of AI, some argue that human review is critical due to trust in LLMs. However, AI can hallucinate and write code that passes tests but fails to meet the intended goals. Manual running of the application and interacting with it provides proof of the pudding test and helps identify issues that automated tools might miss.",
  "summary": "Should we expect human software developers to review code written by AI? For many the answer is clearly yes: how else will we maintain accountability? Few have challenged the thinking, but many of the advantages of past code review practices have evaporated in modern software development. Underneath the question is another one: how do we know code is correct? For a simple function we can…",
  "key_points": [
    "Traditional code review focused on syntax and standards, lacking teaching opportunities.",
    "Pair programming combined various evidence types, providing stronger defect detection.",
    "AI-generated code may lack real-world proof, making manual review critical."
  ],
  "editors_take": "The evolution of code review practices reflects a shift from seeking definitive proof of code correctness to gathering evidence of quality, highlighting the limitations of human review in ensuring accountability.",
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}