We Can't Name Variables. Now We're Writing Prompts? (davidadamojr.com)

AI isn’t just changing how we code — it’s changing what “code” is. The article argues that prompts have become the new source code: unlike compiled languages with type systems and linters that force precision, natural language is ambiguous and won’t throw syntax errors — it will just guess. That shift makes specification and communication the primary bottleneck in software development: logic generation is now cheap (models produce working code fast) but clarity is expensive. Engineers who can precisely externalize intent will get useful output; those who can’t will generate plausible-looking but incorrect systems and then scramble to debug both the code and their vague prompts. Practically, prompting is a discipline akin to naming and API design: prompts are compressed context, ambiguity is the enemy, and iteration is mandatory. The piece gives concrete guidance: state exact constraints (e.g., “reduce time complexity from O(n²) to O(n log n)” or “add retry with exponential backoff, timeout after 30s, log to Sentry”), provide example inputs/expected outputs (TDD for prompts), and build incrementally (scaffold routes, then validation, then caching). The implication for AI/ML teams and hiring is profound: evaluation shifts from raw algorithmic speed to communication, specification, and review skills. Improve prompting by practicing technical writing, better naming, and small-step iteration — treat prompts as precise specs for a literal, fast, slightly confused implementer.