Agentic Design Patterns

Architectural patterns for building agentic AI systems, from Google Cloud’s Architecture Center guidance on choosing between them. Each page covers one pattern: how it works, when to use it, and what it costs.

Start here

  • Choosing an Agentic Design Pattern — the decision framework: whether you need an agent at all, the four requirement dimensions, and which pattern fits which workload shape.

Single agent

  • Single-Agent Pattern — one model, one tool set, one system prompt; the baseline architecture and the recommended starting point.
  • ReAct Pattern — thought, action, observation in a loop; the reasoning technique that raises a single agent’s ceiling.

Multi-agent

  • Multi-Agent Systems — the shared principle behind every pattern below: decompose an objective across specialized agents, and engineer the context each one sees.

Orchestrated by code

Workflow agents that run on predefined logic, without a model deciding what happens next.

  • Sequential Pattern — a fixed linear chain where each agent’s output is the next agent’s input.
  • Parallel Pattern — concurrent fan-out to independent subagents, then a gather step that consolidates their outputs.
  • Loop Pattern — a sequence of subagents repeats until a termination condition is met.
  • Review and Critique Pattern — a loop implementation: a generator produces output, a critic gates it against fixed criteria.
  • Iterative Refinement Pattern — a loop implementation: agents progressively improve a result held in session state.

Orchestrated by a model

Routing decided at runtime by model reasoning.

Special requirements

  • Human-in-the-Loop Pattern — the agent pauses at a checkpoint and waits for a person to approve, correct, or supply input.
  • Custom Logic Pattern — orchestration written as code, mixing the other patterns for workflows that fit no template.

Graph View