The deeper an AI design tool enters real projects, the more rule conflicts it will face.

A designer has personal preferences. A project has explicit requirements. A client gives temporary feedback. The product itself has defaults. Without priority, AI can easily mix these layers together.

My recommended starting order is:

product defaults
-> designer profile
-> current project rules
-> explicit instruction in the current session

Why project rules should beat personal preferences

Personal preferences matter. A designer may have long-term preferences for clearance, material expression, storage strategy, or lighting.

But once the work enters a specific project, project rules should take priority.

The reason is simple: project rules usually come from the client brief, site conditions, budget, construction limits, and specific users. They are not abstract taste. They are the boundaries of the current project.

If a personal profile overrides project rules, AI may force an old habit into a project where it does not fit.

Why current instructions should be highest

An explicit instruction in the current session usually represents a local design decision.

For example, the project may prefer wider circulation, but the designer says: “For this temporary option, allow the passage to become narrower so I can see how much storage improves.”

AI can execute that local instruction, but it must know that this is a session instruction. It is not a permanent project rule, and it is not an update to the designer’s profile.

Current instruction being highest does not mean it can silently rewrite long-term rules.

Every rule should know its source

Priority is important, but provenance is just as important.

AI should be able to explain:

  • did this rule come from product defaults or the designer profile?
  • is it a current project rule or a temporary session instruction?
  • did it override a lower-priority rule?
  • should it be written back to the project?
  • does it need confirmation before being saved long term?

Without source explanation, “following rules” becomes a black box.

Do not turn project exceptions into global habits

Design projects often contain one-off exceptions.

One client prefers a special material. One floor plan has unusual circulation. One source drawing needs a special interpretation. These can be recorded, but they should not automatically become global rules.

A good AI design workbench should separate:

  • product defaults: suitable for all users;
  • designer profile: long-term habits for one designer;
  • project rules: valid only for the current project;
  • session instructions: valid only for the current operation.

Only with this separation can AI avoid polluting future projects with accidental decisions from one project.