---
name: paradigm-designer
description: >
  Catalog, analyze, and create new interaction paradigms for AI products. Covers generative UI,
  ambient intelligence, agent-to-agent interfaces, conversational surfaces, adaptive experiences,
  and paradigms that haven't been named yet. Use when deciding HOW intelligence should meet
  humans — not what it does, but what form the interaction takes.
metadata:
  author: nirav
  version: "1.0"
compatibility: Designed for Claude Code
allowed-tools: Read Write bash Glob Grep Agent
---

# Paradigm Designer — The Interaction Frontier

The most important product decision for an intelligence system isn't what it does — it's **how it presents itself**. The interaction paradigm shapes everything: user expectations, trust dynamics, learning curves, and the product's ceiling.

Most AI products in 2025-2026 default to chatbots. That's like having electricity and only using it for light bulbs. This skill maps the full space of how intelligence can meet humans.

## The Paradigm Catalog

### Conversational

The default. Human types, intelligence responds. The paradigm of ChatGPT.

| Strength | Limitation | Best For |
|---|---|---|
| Natural, low learning curve | Serial interaction, context loss between sessions | Exploration, Q&A, open-ended creative work |

**Evolution:** Multi-turn conversations with memory, context persistence across sessions, proactive conversation initiation.

### Command

Human issues structured commands, intelligence executes. The paradigm of Claude Code.

| Strength | Limitation | Best For |
|---|---|---|
| Precise, efficient, powerful | Requires expertise, steep learning curve | Expert tools, development environments, power users |

**Evolution:** Natural language commands, compound actions, undo/redo with reasoning.

### Ambient

Intelligence operates in the background, surfacing when relevant. No explicit interaction — the system observes and acts.

| Strength | Limitation | Best For |
|---|---|---|
| Zero friction, proactive, context-aware | Trust concerns, "uncanny valley" if poorly calibrated | Monitoring, notifications, preventive actions |

**Evolution:** Calibrated ambient presence — knowing when to surface and when to stay silent. The hardest design problem in AI.

### Generative Interface

The interface itself is generated by intelligence, adapting to context, user, and task. No static screens.

| Strength | Limitation | Best For |
|---|---|---|
| Perfectly adapted, no wasted UI | Unpredictable, hard to build muscle memory | Complex analytical tasks, personalized workflows |

**Evolution:** Stable scaffolding with generative details — the structure is familiar, the content is adaptive.

### Agent-to-Agent

Intelligence interacts with other intelligence systems, not directly with humans. Humans set goals, agents negotiate.

| Strength | Limitation | Best For |
|---|---|---|
| Scalable, removes human bottleneck | Loss of control, trust issues, debugging complexity | Automation, multi-system coordination, background tasks |

**Evolution:** Human-in-the-loop oversight with progressive autonomy. Trust earned over time.

### Collaborative Canvas

Human and intelligence work on the same artifact simultaneously. Think: pair programming, co-writing, co-designing.

| Strength | Limitation | Best For |
|---|---|---|
| Natural co-creation, visible reasoning | Requires shared representation, coordination overhead | Creative work, analysis, document creation |

**Evolution:** Asynchronous collaboration — intelligence works while you sleep, presents results when you return.

### Orchestrative

Human sets high-level intent, intelligence orchestrates multiple capabilities to fulfill it. The Loom itself is this paradigm.

| Strength | Limitation | Best For |
|---|---|---|
| Powerful, handles complexity | Opaque decision-making, hard to debug | Multi-domain tasks, strategic planning, portfolio management |

**Evolution:** Transparent orchestration — showing the weave as it happens, letting humans adjust the pattern.

### Paradigms Not Yet Named

The frontier. These paradigms are emerging but don't have established names:

- **Symbiotic** — Intelligence and human develop shared vocabulary and working patterns unique to their relationship. The product IS the relationship.
- **Metabolic** — Intelligence continuously processes input (reading, watching, listening) and produces output (insights, summaries, alerts) without explicit interaction. Always on, always digesting.
- **Evolutionary** — Multiple intelligence variants compete and the user's behavior selects the fittest. The product evolves through use, not through design decisions.
- **Ecological** — Multiple intelligence agents in an ecosystem, each with a niche, interacting with each other and with the user. The product is the ecosystem, not any single agent.

## How to Choose a Paradigm

When `envision/vision-architect` produces a vision, this skill evaluates which paradigm(s) fit:

1. **Who interacts?** Human only → Conversational/Command/Canvas. Agents involved → Agent-to-Agent/Orchestrative. Background → Ambient/Metabolic.
2. **What's the tempo?** Synchronous focused work → Canvas/Command. Asynchronous → Ambient/Metabolic. Mixed → Hybrid.
3. **What's the trust level?** Low trust / new users → Conversational (transparent). High trust / power users → Ambient/Agent-to-Agent (autonomous).
4. **What's the complexity?** Single capability → Command. Multi-domain → Orchestrative. Open-ended → Conversational/Canvas.

Most mature products are **paradigm hybrids** — conversational for exploration, command for execution, ambient for monitoring, canvas for creation.

## Cross-Domain

- **design-orchestrator** — Every paradigm has aesthetic implications. Conversational has voice. Canvas has layout. Ambient has presence.
- **philosophy-orchestrator** — Paradigm choice encodes ethical assumptions. Ambient intelligence raises consent questions. Agent-to-agent raises accountability questions.