First screen, captured 3 days ago. Hover to scroll the page, or open the video.

The foolishness of "natural language programming" (1978)

What HN thinks

Dijkstra posts from 1978, and the thread spends the day arguing about whether prompting Claude counts as programming or as management with extra steps.

Dijkstra still right: precision needs formalism
Natural language is ambiguous, so real programs still need formal languages. LLMs write the code, but the code, not the prompt, is the program.
LLMs replace developers, not compilers
An agent turning a spec into code is like a human developer doing it, not like a compiler that preserves semantics. Calling English a higher-level language muddles that difference.
Prompting an agent is programming
Agents act as interpreters of natural language. Nondeterminism exists in compilers and JITs too, so the line is blurrier than critics claim.

Consensus: Most commenters agree that deterministic, formal code remains the essential artifact, and that agents rely on compilers and type checkers rather than replacing them. Whether instructing an agent deserves the name "programming" stays disputed.

Recommended in the comments

Inform 7 ↗ (tool): Cited as the most elegant attempt to blend natural language and programming, though some found its loose syntax hard to master.
F* (tool): Suggested as the kind of expressive type system that agents can use to check their own output for semantic correctness.
Playwright (library): Used to have agents write long end-to-end tests that guard against regressions.