The End of Programming

Claims that AI coding agents herald “the end of programming” are drawing both excitement and skepticism. Commenters point to large-scale examples like an AI-assisted rewrite of the Bun JavaScript runtime as proof that models can now generate and refactor huge codebases, but argue this still relies on human-crafted tests, architectures, and product decisions. Many expect the nature of software work to shift toward specification, verification, and system understanding rather than line-by-line coding, while warning that overreliance on AI may erode expertise, degrade software quality, and upend developer job markets.

Overall reaction to “end of programming”

  • Many see the headline as overblown or clickbait; others say it accurately describes a real shift in how code is produced.
  • Historical parallels are drawn (e.g., FORTRAN “ending coding”); consensus is that “programming” changes form rather than disappearing.
  • Several argue this is more about the end of manual coding as default than the end of software engineering.

Bun Zig→Rust rewrite as evidence

  • Supporters see the Bun rewrite as a striking proof that agents can generate and iteratively refine large, complex codebases when there is a strong oracle (tests, reference implementation).
  • Critics counter that:
    • Translation with a full test suite is a best‑case, not representative of typical projects.
    • The resulting Rust code reportedly contained serious soundness and quality issues.
    • Much of the benefit may come from general refactors, not just the language switch or AI itself.
  • Several note we must assume more human review and engineering happened than marketing implies.

Code quality, architecture, and maintenance

  • Repeated reports of LLM output being “slop”: buggy, overconfident, weak at refactoring, and bad at architecture unless tightly guided.
  • People see LLMs excelling at boilerplate, translation, and small tools, but struggling as systems get more complex or older.
  • Strong concern that dependence on AI erodes human expertise, leading to poor architectural decisions and “cognitive debt.”
  • Tests and verification are seen as critical; some believe code will become disposable, regenerated from specs/tests, others worry about endless regressions without near‑perfect coverage.

Changing role of programmers

  • Many expect a shift from line‑by‑line coding to:
    • Defining precise specs, constraints, and verification harnesses.
    • System architecture, trade‑offs, and product decisions.
    • Reviewing, debugging, and maintaining AI‑generated code.
  • Some think this will favor highly skilled engineers and push out “just here for the money” devs; others fear the opposite—loss of craft and passion, replaced by prompt jockeying.

Economic and labor impacts

  • Predictions range from:
    • Smaller onshore teams + big token budgets replacing large engineering groups or outsourcing.
    • Significant deflationary pressure on white‑collar wages globally.
  • Some foresee software skills becoming table‑stakes for product roles, shrinking dedicated “software engineer” headcount.
  • There is anxiety about social readiness: talk of UBI or even “universal high income,” plus worry that current political systems won’t respond adequately.

Long‑term trajectory and open questions

  • Several frame this as the rise of metaprogramming / declarative or intent‑based development, not extinction.
  • Others argue that fully agentic systems deciding rewrites and product direction would be unsafe or economically misaligned.
  • Unclear whether future models can truly handle high‑level product, political, and organizational context, not just code.