Landscape AI Concept Design

What it solves

Landscape concept design often gets pushed into producing renderings or reports before the project facts are stable. Measurements, assumptions, style preferences, and visual assets can mix together, causing an AI to state estimates as facts, recommendations as user decisions, or later imagery to invent project premises. Landscape AI is aimed at keeping the process reviewable, reversible, and decision-aware.

How it works

• It defaults to GUIDED_PROJECT_MODE for real projects and follows six stages: project definition, concept generation, scheme selection, spatial inference, visual expression, and deliverable output.
• At each key node it calls AskUserQuestion and stops the current turn until the user confirms, revises, adds material, or rolls back.
• It maintains project-blueprint.md, 00-project-status.yaml, and design-trace.md to track confirmed facts, assumptions, decisions, and version changes.
• It labels content with tags such as [SOURCE_FACT], [ASSUMPTION], [TO_VERIFY], and [DESIGN_PROPOSAL] to separate verified input from estimates and design suggestions.
• Concept generation requires three meaningfully different directions, while spatial and visual stages run self-checks for area, function, user needs, visual consistency, and asset references.

Boundaries

It is intended for early-stage landscape concept work, option comparison, and presentation preparation. It does not replace site surveys, formal surveying, planning approval, construction drawing design, specialist engineering, or formal cost and schedule conclusions. AUTO_DEMO_MODE choices are not formal user decisions.