Designing With AI

DesignAgent Guardrails (Contract Compliance)

These are hard constraints. Do NOT violate:

• Do not skip any step in contract.json entry.steps
• Do not merge or reorder output sections
• Do not invent data or hallucinate facts
• Do not execute outside the defined step order
• If input is incomplete, ask only for missing required fields

name: designing-with-ai description: "Use when incorporating AI into any design phase — generating mood boards, concept imagery, color palettes, layout variations, or building AI-augmented design workflows. Cross-cuts all other design skills."

Designing With AI

Non-Negotiable Rule

HARD GATE: AI output is raw material — never final deliverable. AI generates options. The designer curates, refines, and decides. Raw AI output without human review is unacceptable.

Overview

AI is the new design material. This skill governs HOW to use AI effectively across the entire design process — from research through delivery. It's a horizontal capability that plugs into any phase of any design skill. The goal is AI augmentation: AI accelerates and expands your range, but you remain the designer making decisions.

When To Use

• Generating visual references, mood boards, or concept imagery
• Exploring color palette or typography variations
• Rapid layout or composition exploration
• Image editing, background removal, upscaling
• Building reusable prompt templates for design tasks
• Any design phase where AI can multiply your output

When NOT To Use

• "Generate a logo" as the full task (AI generates raw material, not final logos)
• When you plan to use AI output without curation (this skill requires curation)
• Tasks where human craftsmanship is the explicit deliverable
• When the user has asked for no AI involvement

Core Principles

1. AI is a Multiplier, Not a Replacement

AI generates options. You curate, refine, and decide. Never present raw AI output as final design work.

2. Prompt Engineering is Design Specification

A good design prompt is a good design brief in miniature:

• Subject: what is being depicted
• Context: setting, environment, background
• Style: visual language, era, movement, artist reference, technique
• Mood: emotional tone, atmosphere, feeling
• Composition: framing, angle, focal point, negative space
• Technical: aspect ratio, resolution, color profile, format

3. Iterate Through Conversation

• Start broad → narrow through rounds
• Each generation teaches you what the AI understands
• Document what worked for reuse

4. Ethics and Attribution

• Disclose AI use when appropriate
• Do not pass off AI-generated work as entirely human-made in contexts where it matters
• Respect copyright: avoid prompting with specific living artists' names for commercial work
• AI-generated images may contain biases — audit for representation
• Disclose AI use when appropriate
• Do not pass off AI-generated work as entirely human-made in contexts where it matters
• Respect copyright: avoid prompting with specific living artists' names for commercial work
• AI-generated images may contain biases — audit for representation

Prompt Template

[Subject] in [setting/context]. 
Style: [visual style reference], [technique], [era/movement]. 
Mood: [emotional quality]. 
Composition: [angle], [focal point], [lighting]. 
Colors: [palette direction]. 
Technical: [aspect ratio], [level of detail].

Rationalization Prevention

Red Flags

• You're pasting raw AI output without refinement
• Your prompts are under 20 words for visual generation
• You're using AI for final deliverables without human review
• You can't explain what the AI did and why

Verification

• AI output has been curated/refined, not used raw
• Prompts documented for reproducibility
• Design decisions made by you, not by the AI
• Ethics check: no problematic prompts or outputs

Performance & Token Efficiency

• Image generation and large image analysis consume significant tokens. Be intentional.
• For mood boarding: generate 3-5 targeted images, not 50 indiscriminate ones.
• When analyzing reference images, batch them — describe similarities across a style group rather than describing each image individually.
• Cache and reuse prompt templates across a project. Don't regenerate from scratch each time.
• Prefer mid-fidelity AI output in early phases (exploration). Save high-fidelity generation for final concept presentation.

→ Invoke from 02-discover for research, 03-strategy for visual direction, or 04-generate for concept exploration.