The "End of Programming" will look a lot like programming (2023)
Claims that AI will bring an imminent “end of programming” draw mixed reactions, with many engineers arguing that current large language models are powerful assistants but far from replacing human developers. Commenters point to enduring challenges around unclear requirements, maintenance, accountability, safety, and real‑world edge cases, likening today’s hype to past waves of UML, no‑code tools, and offshore development that promised to eliminate programmers but mostly changed how they work. The emerging consensus is that AI will increasingly handle boilerplate and routine coding, while humans focus more on specification, testing, architecture, and oversight rather than disappearing from the loop.
Scope of “End of Programming”
- Many argue “no programming” would mean AIs independently handling tasks end-to-end (emails, APIs, hiring other AIs, etc.), with users issuing natural-language requests like “show me my best customers,” no explicit code or UIs.
- Others think the “end” just shifts programming into a half-natural, half-formal language, or into specifying rules, tests, and requirements rather than writing functions.
Current & Near-Term Reality
- Today’s LLMs are seen as:
- Strong autocomplete for verbose languages.
- Faster Q&A than searching/Stack Overflow.
- Good for boilerplate, one-liners, and initial tests, but not complex or novel systems.
- Several report success with agents that can read/write files, iterate, and fix simple bugs (e.g., small games), yet these still struggle with unfamiliar or complex codebases.
Will Programmers Be Replaced?
- Skeptical view:
- Programming difficulty lies in understanding requirements, tradeoffs, and human desires, not syntax.
- History of CASE tools, UML, 4GLs, no-code/low-code: they shifted work rather than eliminated programmers.
- LLMs face “reality vs text” mismatches (e.g., hardware bugs, real-world constraints), so human oversight is seen as permanent.
- More optimistic view:
- As models improve, they’ll handle both coding and maintenance; humans may primarily specify behavior and write tests.
- Developers may evolve into architects, tech leads, or “prompt programmers” managing fleets of AI agents.
Accountability, Testing, and Specs
- Strong emphasis that AI cannot be legally accountable; organizations will still need humans to own results.
- Many foresee programmers becoming testers/spec writers: tests and properties encode the true business logic, while AI generates implementations.
- Writing good requirements is highlighted as at least as hard as coding; plain-English specs will still need precision.
Economic and Organizational Concerns
- Businesses will push AI to cut costs, even when quality suffers, as already seen with outsourcing and some AI-assisted workflows.
- Fears center less on total replacement and more on:
- Staff reduction and pressure on weaker developers.
- Micromanagement and automated performance scoring driven by AI.
- Hybrid environments where humans must maintain AI-generated “hairball” systems with little control over underlying design.