AWS Defines Four Building Blocks Behind Amazon Quick’s AI Agents

A common language for agentic work

AWS says Amazon Quick’s agents have evolved from answering questions to taking actions such as scheduling meetings, running research and generating reports. During that development, the company found that similar concepts were repeatedly being built under different names.

The taxonomy is intended to clarify requests that could otherwise be interpreted in several ways. For example, a request to monitor a Slack channel and send a daily digest might be treated as a new agent, a scheduled instruction, a workflow or a skill. AWS’s framework places these elements into separate roles, while allowing them to work together.

According to AWS, a trigger starts a task and carries its instructions. An agent takes responsibility for the work, while capabilities determine how it is performed. The company says most of this framework should remain invisible to end users, even though it gives developers a structure for reasoning about the system.

Triggers define when work starts

AWS identifies four trigger types. Human-initiated triggers begin when a person asks an agent to do something in a conversation. Agent-delegated triggers allow one agent to invoke another for a sub-problem. Scheduled triggers activate work at a specified time, while event-driven triggers respond to signals such as a new email, a Slack mention, a file change or a webhook.

The trigger type can influence execution. AWS gives the example of a scheduled background task staying conservative and doing nothing when uncertain, while a human-initiated task might ask a follow-up question instead.

The company defines an agent as a “persistent, learning actor.” Each agent includes standing instructions, available capabilities and learned context. AWS says the underlying large language model is a swappable parameter, conversation history belongs to the session, and triggers belong to the platform.

Tools, skills and workflows set the confidence contract

Capabilities describe how an agent acts, ranging from flexible, agent-led decisions to deterministic execution. Tools are atomic operations with typed inputs and outputs, such as sending an email, querying a database or reading a file. Skills provide structured guidance, resources and examples while leaving room for judgment. AWS says skills use progressive disclosure, showing short descriptions until the agent needs the full instructions.

Workflows coordinate declared steps, branching and error handling in a fixed order, with decision logic in code rather than model inference. AWS describes this as useful for processes such as purchase approval, where a budget check, manager sign-off and finance review can be recorded.

The company argues that skills and workflows retain value as models improve. Skills capture, version and share institutional expertise, while workflows provide auditability and deterministic coordination. AWS says a skill can become a workflow when its process is well understood, while a workflow can loosen into a skill when it repeatedly encounters edge cases.

For users, the practical message is that the same daily Slack digest could begin as a simple scheduled instruction, gain a skill with formatting rules and priorities, or use a workflow with fixed steps and error handling. AWS says the appropriate choice depends on how much latitude the user is willing to give the agent, with learning potentially improving later runs without manually rewriting the instructions.