Taxonomy of agentic primitives in Amazon Quick
As we built Amazon Quick, its agents grew from answering questions to taking actions, and the same primitives kept getting rebuilt under different names. This post shares the taxonomy we converged on — triggers, agents, capabilities, and learned context — found by decomposing agentic work into four plain questions: When, What, Who, and How.
Over the past year, Amazon Quick agents transitioned from merely answering questions to executing actions like scheduling meetings and generating reports. However, a recurring issue emerged - there was no standardized definition for the core components that make up agentic work. Different concepts were repeatedly built under varying names such as actors, signals, procedures and memory.
This led to confusion during product design and user interaction. For instance, the user's request "monitor my Slack channel and send me a daily digest" could be interpreted as a new agent, a scheduled instruction, a workflow, or a skill - all different interpretations of the same query. To address this, the article introduces a taxonomy of agentic primitives: triggers, agents, and capabilities.
The four fundamental questions used to decompose agentic work are When?, What?, Who? and How?. Triggers initiate the work and carry its instructions, while agents serve as the "Who?" executing the work and leveraging their capabilities (tools, skills and workflows) to carry out the task (How?). The article further breaks down triggers as human-initiated (conversational), agent-delegated (one agent calling another), scheduled (time-based) and event-driven (signals from external systems).
Agents are defined as persistent learning actors who interpret instructions, carry capabilities and accumulate memory through use. These agents can be human-initiated, agent-delegated, scheduled or event-driven. Capabilities are the mechanisms through which agents act, ranging from fully agentic to fully deterministic. This includes tools (atomic operations with typed schemas), skills (structured guidance with decision-making capability) and workflows (deterministic coordination of steps with fixed execution order).
The choice between skills and workflows depends on whether the agent or the user determines the step order. The article concludes that despite ongoing model improvements, skills and workflows remain valuable for capturing and sharing institutional expertise, providing auditability and visibility, and enabling version control. The taxonomy aims to provide a shared vocabulary for anyone building on top of Amazon Quick, making the system easier to reason about.
Written by urgent.news from AWS Machine Learning's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
