The Principle × Lifecycle Matrix
17 principles down the side, seven lifecycle stages across the top. Each cell is an intersection — tap a chip to open the pattern.
Listen to this page
The Agentic AI Pattern Catalogue
This page shows a big grid, almost like a bingo board, that connects two ideas: something an AI agent should always do (called a principle, like "stay safe" or "keep good records") and the moment in its work when that idea matters most (called a stage, like "planning" or "taking action"). Each box where a row and column meet can hold one or more patterns that fit both ideas.
This matters because it helps you find the right pattern fast instead of guessing. If you know what stage your AI agent is at and what you are worried about, you can look at that one box and see exactly which patterns to use. Switch between two different views of the agent's lifecycle with the toggle above, and click any colored tag to open that pattern's full page.
| Principle | 01Perceive | 02Reason / Plan | 03Decide | 04Act | 05Observe | 06Reflect / Learn | 07Terminate |
|---|---|---|---|---|---|---|---|
| F-01Bounded Autonomy | |||||||
| F-02Goal Alignment & Scoping | |||||||
| F-03Governance-by-Design | |||||||
| A-01Perceive–Reason–Act Cycle | |||||||
| A-02Reasoning & World Model | |||||||
| A-03Memory Stratification | |||||||
| A-04Tool Use as Typed Contract | |||||||
| A-05Planning & Decomposition | |||||||
| A-06Knowledge Grounding (RAG) | |||||||
| A-07Inter-Agent Communication | |||||||
| A-08Orchestration Topology | |||||||
| O-01Human-in-the-Loop Checkpoints | |||||||
| O-02Observability & Traceability | |||||||
| O-03Guardrails & Safety Layers | |||||||
| O-04Evaluation & Continuous Monitoring | |||||||
| O-05Exception Handling & Graceful Degradation | |||||||
| O-06Identity & Least-Privilege |
What Do the Seven Stages Mean?
The column headers above are short. Here is what each one means in plain words, for both lifecycle views.
Runtime Cognitive
| # | Stage | What It Means |
|---|---|---|
| 01 | Perceive | The agent looks around and takes in new information, like reading a message or a file. |
| 02 | Reason / Plan | The agent thinks about the goal and breaks the big job into small steps. |
| 03 | Decide | The agent picks what to do next, like choosing which tool to use. |
| 04 | Act | The agent does the thing — it uses a tool or makes a change. |
| 05 | Observe | The agent checks what happened after it acted, and learns the new result. |
| 06 | Reflect / Learn | The agent looks back at its own work, checks for mistakes, and remembers lessons for later. |
| 07 | Terminate | The agent stops. It may finish the job, ask a person for help, or hand the work to someone else. |
AgentOps Lifecycle
| # | Stage | What It Means |
|---|---|---|
| 01 | Design & Scope | The team decides what the agent may do, and how much freedom it gets. |
| 02 | Build & Integrate | The team picks the AI model, connects the tools it needs, and plans how it remembers things. |
| 03 | Evaluate & Red-Team | The team tests the agent hard, trying to break it, to make sure it is safe before it goes live. |
| 04 | Deploy | The agent is released slowly and carefully, often in a safe practice space first. |
| 05 | Operate & Monitor | The team watches the agent while it works, checking that it behaves well. |
| 06 | Iterate & Adapt | The team keeps improving the agent, fixing its instructions over time. |
| 07 | Decommission | When the agent is no longer needed, the team turns it off safely and clears its data. |
