# Human-AI Collaboration Patterns The spectrum of how humans and AI work together. [[Human-in-the-Loop]] is one pattern (an approval gate), but the full spectrum is wider. Five distinct patterns exist, each with different trust, risk, and context requirements: 1. **AI assists** (human drives, AI helps). The human does the work; AI provides suggestions, autocomplete, or lookups on demand. Lowest risk, lowest leverage. Example: code completion, spell check. 2. **AI suggests** (AI proposes, human decides). AI generates options or recommendations; the human evaluates and chooses. Moderate leverage. Example: AI suggests meeting times, draft responses, research angles. 3. **AI drafts** (AI creates, human refines). AI produces a first version; the human edits, reshapes, and approves. High leverage when the human has strong editorial judgment. Example: AI drafts an article, the human rewrites it. 4. **AI executes with oversight** (AI acts, human monitors). AI performs tasks autonomously but within guardrails, with human review at checkpoints. Requires [[AI Agent Permissions]] and clear boundaries. Example: AI processes email, files documents, updates databases. 5. **AI autonomous** (AI acts independently). AI operates without human intervention within a defined scope. Highest leverage, highest risk. Requires robust [[AI Safety]] measures and the trust to match. Example: automated trading, self-healing infrastructure. Each pattern maps to different [[Levels of AI use]]. Moving up the spectrum requires progressively more trust, better context, and stronger guardrails. The [[Lethal Trifecta for AI Agents]] describes what goes wrong when autonomy is granted without adequate safeguards. The right pattern depends on the task, not the technology. High-stakes decisions should stay lower on the spectrum. Routine, reversible tasks can safely move higher. ## References ## Related - [[Human-in-the-Loop]] - [[AI Agents]] - [[AI Agent Permissions]] - [[Levels of AI use]] - [[AI Safety]] - [[Lethal Trifecta for AI Agents]]