You Cannot Outsource Understanding (russmiles.substack.com)

This essay argues that attempts to eliminate developers—via outsourcing, no-code, or AI code generation—misunderstand what software development actually buys: contextual understanding, not just production. History repeats: offshoring lost domain empathy, no-code created brittle automations, and now AI promises velocity but increases code volume and cognitive load (a software Jevons’ Paradox). Every line of generated code still needs someone to test, integrate, observe, and evolve. For AI/ML teams this matters because model-infused systems amplify complexity: data, model behavior, pipelines, and infra all require continuous, domain-informed reasoning that tools alone cannot supply. Practically, the piece urges designing platforms and practices that amplify developer understanding rather than replace it: pair developers with domain experts, give platform teams responsibility for enabling comprehension (observability, self‑service docs, explainable tooling), and measure learning velocity not just delivery speed. Avoid mistaking efficiency for comprehension, treating developers as interchangeable, or believing abstraction removes complexity. For AI/ML practitioners, that means treating codegen and copilots as accelerators of insight, investing in context ownership, and building tooling to read and explore system behavior—not just write it—so teams can safely scale model-driven systems without outsourcing the one thing machines can’t automate: human understanding.