Should I design for humans or machines?

Where human-centered design breaks down

Human-centered patterns rely on signifiers, affordances, and shared context. A designer's note like "use this variant for emphasis" works for people who understand visual hierarchy, but it's useless to a system that must decide programmatically when to apply a variant. Machines need determinism: explicit conditions, consistent structure, and rules they can evaluate.

Three concrete changes to documentation

• Define inputs and outputs. Instead of vague guidance, documentation should state the exact triggers that determine a component's appearance or behavior. Example: map priority levels or message types to specific variants.

• Reconstruct narrative as structured data. Break prose into design tokens, component properties, and machine-readable fields so tooling can apply intent automatically.

Design systems were built to be read by people. Their examples, guidelines, and rationales live in natural language, Figma comments, and engineers' heads. Those formats are brittle when consumed by automation. To support machine users, a system's documentation must be explicit and consistent enough to be parsed and executed without requiring subjective judgment.

Practical implications for everyday work

Designers now operate in two overlapping worlds: one that prioritizes human intuition and one that demands machine precision. That affects how you write guidelines, model components, and collaborate with engineers. Expect to:

• Turn ambiguous rules into concrete properties and tokenized values.
• Add decision logic to component docs so automation has deterministic rules to follow.
• Audit legacy guidance for machine-readability — look for vague phrases like "use when appropriate" or "avoid in complex flows." Those need definition.

Teams that adopt structured, rule-based documentation will get more reliable automation and fewer edge cases when agents generate UI or wire up components. Teams that keep relying on flexible, human-only guidance will find automation failing or producing inconsistent results. The design trade-off is real: preserving human nuance while providing the clarity machines require.

Designers should not choose between humans or machines. Instead, produce guidance that serves both: keep human-facing explanations for context and decision-making, and also publish the structured rules, tokens, and conditional logic machines need. That dual approach preserves human judgment where it matters and makes automation predictable where it matters.