30 · Choosing the Right Agent Architecture

This is a decision framework for choosing the right agent architecture: start simple and add complexity only when it creates real value. Use it to pick the right pattern, stack and services.

How it works

1. 1Can a single LLM answer directly? Use a prompt.
2. 2Does it need external information? Add RAG.
3. 3Is it a fixed sequence of steps? Use a workflow (LangGraph).
4. 4Are there independent subtasks? Use parallel agents.
5. 5Open-ended, long-horizon, continuous adaptation? Use autonomous agents with AgentCore.

Key points

• Pattern summary maps each pattern to its best use case and LangGraph mapping.
• Consider business value, complexity, latency, cost, risk and governance before choosing.
• Anti-patterns to avoid: multi-agent when one works, over-engineering, ignoring failure modes.
• Right architecture + right pattern + right stack = agents that deliver real impact.

Why it matters

Most failures come from choosing the wrong complexity. A decision framework keeps you on the simplest effective architecture and away from costly over-engineering.

Frequently asked questions

What's the golden rule?

Start with the simplest thing that works, measure, then evolve — don't reach for multi-agent by default.

When do I need autonomous agents?

For open-ended, long-horizon goals that need continuous planning, memory and adaptation.