Coding in English (www.nvegater.com)

AI coding today hands “superpowers” to non-developers by turning English into working implementations, but often only marginally speeds up experienced engineers. The piece argues the real opportunity isn’t replacing code with prose, but building tools that tightly link natural-language specs and code: a spec-to-code compiler that flags ambiguous sentences, detects conflicts with existing logic, traces each sentence to the code it generates, and lets edits flow both directions so code changes suggest spec updates. That would smooth the long-term trend of raising abstraction while preserving the precision needed for reliable software. For the AI/ML community this is a meaningful roadmap: it demands advances in semantic parsing, program synthesis, provenance/traceability, ambiguity detection, and verification. Practically, the system would need live feedback like an LSP, conflict resolution strategies for under-specified requirements, and mechanisms to maintain sync between human intent and generated implementations. The idea surfaces challenging research problems (formalizing spec semantics, round-trip editing, test-generation from prose) and attractive product implications—broader participation in software development and more robust natural-language programming—making it a ripe area for both applied ML teams and tools startups.