AWS Amazon Quick Team Introduces Taxonomy of Agentic Primitives
AI assistants · AI coding
AWS team members behind Amazon Quick outlined a conceptual framework organizing AI agent architectures around four basic primitives: triggers, agents, capabilities, and learned context.
The Amazon Quick engineering team shared an architectural taxonomy designed to formalize primitives in agentic software systems. The model breaks task execution down into four structural questions: when, what, who, and how. In this model, triggers specify the activation event and task instructions, agents represent persistent learning entities comprising standing instructions, capabilities, and accumulated context, and capabilities define how actions are taken across varying levels of determinism using tools, skills, and code-based workflows.
Key takeaways
- Triggers combine activation timing (When) and task instruction (What), categorized as human-initiated, agent-delegated, scheduled, or event-driven.
- An agent is defined as a persistent learning actor (Who) structured by instructions, capabilities, and learned context, rather than the underlying swappable model.
- Capabilities (How) span tools (atomic operations), skills (guidance loaded via progressive disclosure), and workflows (deterministic code execution).
Source：AWS Machine Learning Read the original ↗
Prepared, checked and automatically published by AI; not individually reviewed by a human.