AI agent cannot create separate code layers for design alternatives

While working in Figma Make, I asked the AI agent to create several alternative versions of an invitation modal as separate code layers and place them below the selected layer.

Earlier, a similar prompt—“set up 3 alternatives”—successfully resulted in separate code layers. However, in the current conversation, the agent instead added all five experiments inside the existing app. After I clarified the request, it said it could not create, duplicate, rename, or position code layers.

The agent inspected the project history and found a Figma-authored “Duplicate code layer” event. It explained that the earlier duplication was performed by Figma Make’s surrounding interface or orchestration layer, while the current agent session only had access to the selected layer’s code repository.

This creates an inconsistent experience: the same type of natural-language request can sometimes produce separate code layers, but at other times the agent cannot access that capability or hand the request back to the appropriate orchestration workflow.

Expected behavior:

• Requests for multiple alternatives should consistently create separate code layers.
• If the current agent cannot perform the operation, Figma Make should route the request to the capability that can, rather than embedding all alternatives inside one app.
• The agent/interface should clearly indicate whether a prompt is operating on one code layer or creating multiple alternatives.