AWS outlines agentic primitives taxonomy for Amazon Quick
A new framework from AWS decomposes agentic work into triggers, agents, capabilities, and learned context to standardize how autonomous systems are built and reasoned about.
1 source · cross-referenced
- Amazon Quick’s agents evolved from answering questions to taking actions, prompting AWS to define a shared taxonomy of agentic primitives.
- The framework centers on four concepts: triggers (When/What), agents (Who), capabilities (How), and learned context (feedback loop).
- AWS argues a common vocabulary reduces confusion in product design and user interaction for complex agentic workflows.
As Amazon Quick’s agents expanded from question-answering to performing actions like scheduling meetings and generating reports, AWS engineers observed recurring patterns in how work was initiated, assigned, and executed. Repeatedly, the same concepts were being rebuilt under different names, creating confusion in product design and user interfaces. To address this, the team formalized a shared taxonomy of agentic primitives grounded in four questions: When, What, Who, and How.
The framework defines triggers as the activation mechanism that both starts work and carries the task description, answering both the When and What of a task. Triggers can be human-initiated (conversational), agent-delegated (sub-task handoffs), scheduled (time-based), or event-driven (external signals like Slack mentions or file changes). Each trigger type conditions execution behavior, with scheduled tasks often prioritizing conservatism and human-initiated tasks favoring responsiveness.
An agent is defined as a persistent, learning actor — the Who in any execution — serving as the anchor for learning and instructions. Agents interpret standing guidance (persona, role, context) and carry histories that differentiate their behavior even under identical inputs. They are distinct from the instructions they follow, emphasizing their role as interpreters of context rather than static rule-followers.
Capabilities represent the tools, skills, and workflows an agent uses to perform tasks, addressing the How of execution. These include external integrations (e.g., Slack, email) and internal procedures, enabling agents to act beyond simple text generation. The taxonomy positions capabilities as modular components that agents can invoke based on the task at hand.
Learned context closes the loop by feeding outcomes back into the system, allowing agents to adapt over time. This feedback mechanism supports continuous improvement, enabling agents to refine their behavior based on session runtime outcomes. AWS frames this as essential for scaling agentic systems beyond static workflows.
AWS describes this taxonomy as foundational to its broader agentic operating system, with future posts planned to detail background learning, proactive feeds, and additional system components. The company positions the framework as a step toward reducing duplication in agentic tooling and clarifying user-facing abstractions for enterprise automation.
- Nov 11, 2026 · Latent Space — swyx
OpenAI demonstrates voice steering, cloud handoffs, and personal agents for Codex at AI Engineer Code Summit
Trust79 - Nov 11, 2026 · The Verge — AI
OpenAI to launch verified AI Business Agents in ChatGPT for US e-commerce and service businesses
Trust79 - Nov 10, 2026 · Latent Space — swyx
NaiveAI releases Naive-N0.5-Flash coding model with ultrafast token throughput
Trust74