Writing Is Closer to Thought Than Code

AI projects often fail not because models cannot execute, but because people have not clearly expressed what they want. Daniel Miessler argues that written intent should be the source of truth, rather than the code an AI produces. Code is a precise implementation, but it is only a representation of the original idea; writing is a closer, more accurate expression of human thought. When a result misses the mark, the remedy is to improve the articulation, not simply to iterate on or document the flawed implementation. The process should have two feedback loops: refine the written vision as understanding develops, then use it to guide and assess the implementation. Building can reveal new preferences, and rereading the vision can change them, so neither needs to be fixed from the outset.

Miessler’s Ideal State Artifact (ISA) puts this approach into practice. It records a prose description of the desired outcome, constraints, dated decisions, and atomic Ideal State Criteria (ISCs), each paired with an evaluation method. The criteria serve both as the build plan and as the test harness, with checks ranging from simple assertions to model-based judgments. Examples include a family meal planner with requirements such as avoiding peanuts and keeping weeknight cooking under 40 minutes, and an owl drawing assessed against explicit visual and emotional criteria. The same principle applies to software: without a written target, accidental choices in an early version can become entrenched as requirements. The ISA keeps revisions anchored to what is actually wanted.