Taxonomy of agentic primitives in Amazon Quick
Amazon Quick introduces a taxonomy that breaks agentic work into four core questions—When?, What?, Who?, and How?—and defines three primary primitives: Triggers (activation mechanisms), Agents (persistent learning actors), and Capabilities (tools, skills, and workflows). This shared vocabulary clarifies distinctions between similar concepts like agents, scheduled instructions, workflows, and skills, making system design more consistent. Engineers can leverage the taxonomy to structure agent interactions, trigger handling, and capability management in their own platforms.
⚡ Key Takeaways
- The taxonomy decomposes agentic work into four plain questions: When?, What?, Who?, and How?.
- Triggers serve as the activation mechanism and carry both the timing (When) and the task description (What).
- Agents are defined as persistent, learning actors that encapsulate instructions, capabilities, and learned context.
- Capabilities encompass tools, skills, workflows, and sub‑agents that can be added or removed dynamically during a session.
- Learning feeds back from the session runtime into the agent’s memory, influencing future executions.
By adopting this explicit primitive taxonomy, engineers can build more modular, maintainable agent systems and avoid ambiguous product designs that blur the lines between agents, schedules, and workflows.
✅ Practical Steps
- Structure your agent system around the four questions (When, What, Who, How) to clearly separate trigger handling, agent identity, and capability execution.
- Implement Triggers that encapsulate both activation timing and task description, supporting human‑initiated, agent‑delegated, scheduled, and event‑driven modes.
- Design Agents to maintain persistent learned context and dynamically manage their Capabilities (tools, skills, workflows) per session.
Want the full story? Read the original article.
Read on AWS ML Blog ↗