The Agentic AI Pattern Catalogue
Approximately sixty composable, technology-agnostic design patterns for autonomous AI agents: mapped to seventeen foundational, architectural, and operational principles across two viewable lifecycles: the runtime cognitive loop and the AgentOps development lifecycle, with comprehensive regulatory standards tracing.
Listen to this catalogue
The Master Blueprint for Agentic AI
A compiled reference, not original research. This microsite curates and reorganises a reference catalogue of agentic AI design patterns, synthesised from primary sources including Anthropic, Antonio Gulli / Google, arXiv pre-prints, AWS, Google Cloud, and Singapore's IMDA Model Governance Framework for Agentic AI. Full attribution and source list on the Standards & Sources page. It is not a standard, and standards-mesh mappings are indicative, not legal advice.
This page is the starting point for a big collection of building plans for AI "agents": smart computer programs that can think, plan, and act on their own, kind of like a robot helper. Each plan is called a pattern, and there are 61 of them, all written down in the same clear way. This matters because AI agents built without a plan can make mistakes, act unsafely, or do things nobody can explain. These patterns are tested ideas that help engineers build agents people can trust.
You do not need to read everything at once. Pick one of four paths: see a big chart of all the patterns at once (Matrix), read short summaries grouped by the companies and researchers who invented them (Field Guide), look up full details for one pattern at a time (Deep Catalogue), or check the safety rules each pattern follows (Standards & Sources).
3Foundational
Principles that define what the agent is: autonomy, goal alignment, governance-by-design.
8Architectural
Principles that define how it is built: perception, reasoning, memory, tools, planning, orchestration.
6Operational
Principles that define how it runs in production, oversight, observability, guardrails, evaluation.
Four ways into the same catalogue.
Start shallow with the Field Guide, or go deep on a single pattern. Every view cross-references the same underlying data.
One pattern and its four coordinates
Worked example from the catalogue data: AUT-DSN-01, Autonomy Tier Classification. Every one of the 61 patterns carries the same four coordinates, which is how the Matrix, Deep Catalogue and Standards views cross-reference one another.
01 · MatrixThe Principle × Lifecycle Matrix
Every pattern sits at the intersection of a governing principle and a lifecycle stage. Toggle between the runtime cognitive loop and the AgentOps development lifecycle to see where each pattern applies.
Open →
02 · Field GuidePattern Field Guide
A clean, source-grouped reference to 91 patterns as originally described by Anthropic, Google Cloud, Gulli, and other primary sources: one paragraph, a use case, and the trade-offs for each.
Open →
03 · Deep CatalogueThe Deep Pattern Catalogue
61 composable patterns treated in full: context, forces, solution, applicability, anti-patterns, consequences, and the regulatory standards each one helps satisfy.
Open →
04 · Standards & SourcesStandards Reference & Bibliography
The thirteen regulatory frameworks every pattern traces to — IMDA MGF, EU AI Act, NIST AI RMF, ISO/IEC 42001, and more — plus the full bibliography and method notes.
Open →Software architecture matured once its patterns were codified. Data architecture is following the same arc. Agentic AI is at the earliest phase: patterns are described variously across blog posts, vendor docs, one book, and a handful of arXiv pre-prints.
This catalogue treats each design decision as a reusable primitive, a named solution to a recurring problem in a specific context, rather than as a framework feature. Reflection is a pattern; LangGraph is an implementation.
This catalogue took longer to build than almost anything else on this site, because I wanted every pattern traced back to where it came from: Anthropic, Google Cloud, IMDA's framework, not paraphrased secondhand. I keep coming back to it myself when a client asks how a specific guardrail should work, because it's faster than re-deriving the answer from scratch. Start with the Field Guide if 61 patterns feels like a lot, it's meant to be read in pieces, not one sitting.
