From the source
Amazon is sharing the taxonomy of agentic primitives it converged on while building Amazon Quick, after finding that the same few ideas kept getting rebuilt under different names.
- The framework decomposes agentic work into four plain questions: When, What, Who and How, with a trigger starting the work and carrying the task, an agent taking it on, and capabilities performing it.
- Amazon defines an agent as a persistent, learning actor made of instructions, capabilities and learned context, and says the underlying large language model is a swappable model parameter while conversation history is per-session state.
- Capabilities span a spectrum from fully agentic to fully deterministic, which Amazon calls a confidence contract: tools are atomic operations with a typed schema, skills are structured guidance with room for judgment, and workflows are deterministic coordination with decision logic in code.
- Amazon says a request such as monitoring Slack and sending a daily digest is usually a trigger added to an existing agent, not a new agent or workflow.
- The post is the first in a series, with upcoming posts on the agentic operating system, background learning and proactive feeds.






