Zenn essay: five ways programmers use AI without writing code

The essay starts from a deliberate constraint: none of the five methods adds a finished program. That separates it from the usual framing of AI for programmers as code generation, design assistance, test creation, or multi-agent management. The author is a programmer writing on Zenn, and the argument is that habits like conditionals, diffs, state, dependencies, and counterexamples can be turned toward one's own decisions and workplace exchanges.

A recurring pattern across the five sections is a two-step discipline. First the AI produces comparison material — a pair of examples differing by one condition, a list of what a document does and does not establish, incompatible rule combinations, or an option that removes work. Then the reader checks that material against the original record or the actual person. The essay repeatedly warns against treating AI agreement as evidence: it cites research on sycophantic responses but explicitly does not claim that rate applies to all current models.

The references are framed as background, not validation. Anthropic's 2023 sycophancy paper, Google's Postmortem Culture, and a 2021 Nature paper on overlooked subtractive changes are listed as related ideas, and the essay states that its two-stage use of AI is its own proposal rather than an official procedure. That framing keeps the piece in the territory of a practitioner's method note, with its claims scoped to what the author proposes rather than to measured results.