{
  "id": 15934121,
  "title": "The compiler was supposed to end programming too",
  "url": "https://urgent.news/2026/10/21/the-compiler-was-supposed-to-end-programming-too",
  "topic": "tech",
  "section": "Tech",
  "published": "2026-10-21T18:00:51.000Z",
  "source": {
    "name": "Hacker News",
    "slug": "hacker-news",
    "url": "https://hrih.si/articles/the-compiler-was-supposed-to-end-programming/"
  },
  "original_language": "en",
  "account": "The compiler was supposed to end programming too. A Tuesday morning in 2031 begins with a programmer having not written any code yet. Overnight agents, including one running on the reporter's own machine, handle the backlog of code. These agents create pull requests with tests, recordings of the features working, and notes about parts that were uncertain. One agent stopped and asked a question about refunds for expired cards, recognizing the need for human judgment.\n\nThis scenario is not science fiction; unattended agents exist today, and one runs on the reporter's machine every night. It invented a small project, built it, tested it, and the reporter reviewed it over breakfast. This project, Rule Safari, is a live project on the reporter's site, with an article explaining its workings.\n\nPeople have been asking if this marks the end of programming for over seventy years. In 1954, IBM's Programming Research Group released a report on FORTRAN, a new language that should virtually eliminate coding and debugging. The industry called this automatic programming. FORTRAN, released in 1957, did eliminate a tedious part of coding, but debugging still existed. It simply moved the job up a level, from registers to larger problems. Programs became cheaper to write, leading to more of them being created.\n\nHowever, this rung is different due to speed. GitHub's 2025 report counted over 180 million developers, with a steep adoption curve of AI tools in coding. From 70% in 2023 to 84% in 2025, nearly half of professional developers use AI tools daily. Favorable sentiment towards AI has declined from 77% in 2023 to 60% in 2025, with 46% distrusting AI output's accuracy and 66% frustrated by nearly correct AI solutions. Debugging AI-generated code takes longer for 45% of developers.\n\nEarly studies, like METR's randomized controlled trial in early 2025, showed a 20% time reduction, but only 19% slower with AI. The study's reliability was questioned due to developers declining to participate without AI assistance. This complex situation raises questions about the role of AI in programming.",
  "summary": null,
  "key_points": [],
  "editors_take": null,
  "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."
}