What should you design yourself, and what should AI do?

Keep ownership of the decisions that define the experience. Delegate work when the output is bounded and you can verify it efficiently. The dividing line will change with the project, but accountability should remain clear even when most of the first draft comes from AI.

Keep human ownership of intent

A designer should frame the problem, decide which people are served, interpret research, choose tradeoffs and approve the final experience. AI can challenge each of those decisions, but it does not carry the relationship with the client or the consequences for users. Define the owner before delegating work.

Handle consequential product rules explicitly. Consent, pricing commitments, cancellation, account recovery and permission boundaries deserve careful human judgment. A plausible default can be wrong for the business and harmful to the person using it.

Delegate bounded production and exploration

Good candidates include drafting content variants from approved facts, populating fictional test data, checking naming consistency, producing repetitive states and proposing alternatives within a defined system. Use a script rather than a language model when the task is a deterministic transformation that a script can perform reliably.

For creative work, delegate breadth before judgment. Ask for different compositions or interaction approaches, then select the direction yourself. A long list of nearly identical options is not productive exploration.

Use a four-question decision rule

Ask: how costly is a mistake, how ambiguous is the task, how important is original authorship, and how easily can the result be checked? A low-consequence, easily checked formatting task is a strong automation candidate. A sensitive research interpretation with incomplete context needs substantial human involvement.

Reconsider delegation when review takes longer than doing the task. If you repeatedly repair the same output, improve the input, narrow the scope or use a more appropriate tool. Automation that creates hidden cleanup is not a capacity gain.

Keep hands-on craft where it shapes character

Design the typographic relationships, signature compositions, illustration direction and interaction details that make the work specific. You may use AI during those activities, but invest attention where small choices carry meaning. Manually refining a logo or a critical interaction can be worth more than generating another hundred variations.

At the same time, manual execution is not automatically good design. A hand-crafted pattern can still be inaccessible or confusing. Judge both human and generated work against the same user goals and quality standards.

Make the split visible to collaborators

In the project log, identify AI-assisted steps, reviewed outputs and unresolved questions. The client does not need a transcript of every prompt; they do need to understand the method when it affects confidentiality, authorship, evidence or maintainability.

A practical handoff says who owns the brief, what the assistant produced, how it was checked and what requires approval. That is a working division of labor rather than a debate about whether AI should be used at all.