Agentic AI Architecture

An agentic AI architecture is the structural answer to "what kind of system is this?". An LLM by itself is a generator; an agent is a generator that decides what to do next, holds state, talks to tools, hands off to other agents, and is observable enough that humans can intervene. The architecture is the set of commitments that lets all of those things happen without colliding.

These commitments come from patterns, not from a single diagram. A supervisor over orchestrator-workers is not the same architecture as a swarm; a ReAct loop with a step budget and decision log is not the same as plan-then-execute. The catalog gives you the vocabulary to make that distinction precise — and to defend it when a stakeholder asks "why did you build it this way?"

Field-tested patterns to start with

• Supervisor — Place a coordinating agent above a set of specialised agents and route work to them.
• Orchestrator-Workers — An orchestrator dynamically breaks a task into subtasks at runtime and delegates each to a worker LLM, then synthesises results.
• Augmented LLM — Build the foundational agent block as an LLM augmented with retrieval, tools, and memory that the model actively chooses to use, rather than a bare-model call.
• ReAct — Interleave a single thought, a single tool call, and a single observation per step so the agent reasons over fresh evidence.
• Plan-and-Execute — Plan all the steps once with a strong model, then execute each step with a cheaper model under the plan.
• Handoff — Transfer the active conversation from one agent to another, carrying context across the switch.
• Subagent Isolation — Run subagents in isolated workspaces so their writes do not collide and parallelism is safe.
• Decision Log — Persist the agent's reasoning trace alongside its actions so post-hoc review can explain why.
• Agent Resumption — Persist agent execution state so a long-running run survives restarts, deploys, or user disconnects.
• Step Budget — Cap the number of tool calls or loop iterations the agent is allowed within a single request.

Recommended reading

• Planning & Control Flow — 40 patterns
• Multi-Agent — 44 patterns
• Governance & Observability — 27 patterns
• Tool Use & Environment — 34 patterns

Or open the full contents for all 421 patterns in 14 books.

Related guides

• LLM Agent Design Patterns — A GoF-formal catalog of LLM agent design patterns: ReAct, tool use, plan-and-execute, reflection, step budget, and more. Each pattern decom…
• RAG Agent Patterns — Patterns for building retrieval-augmented generation agents: naive RAG, agentic RAG, hybrid search, cross-encoder reranking, contextual ret…
• Multi-Agent Patterns — Patterns for coordinating multiple LLM agents: supervisor, orchestrator-workers, handoff, debate, hierarchical agents, swarm, role assignme…
• AI Agent Safety Patterns — Safety patterns for LLM agents: step budget, kill switch, constitutional charter, approval queue, sandbox isolation, input/output guardrail…