Practical discipline for working with LLMs: self-review and judgment

Short summary

An opinion piece arguing that most LLM users accept first outputs without critical review, when simply asking the model to self-check catches cheap errors. The author frames engineering judgment — choosing the right solution among ten options — as the irreplaceable human layer. Cosmetic code-style debates are dismissed as hobbyism now solvable with linter configs.

• •Ask LLMs to self-review outputs before accepting them as final
• •Engineering judgment in selecting the right solution remains the human core
• •Cosmetic style preferences should be encoded in tooling, not debated manually

Generated with AI, which can make mistakes.