Amazon's Strands Agents SDK Cuts Agent Setup from 40 Lines to 3
AWS open-sourced Strands Agents, a model-driven SDK that runs production agents across any provider with hooks, steering, and built-in observability.
PRO- AWS open-sourced Strands Agents, a Python and TypeScript SDK for production AI agents.
- Model-driven loop works with Bedrock, Anthropic, OpenAI, Gemini, Ollama, and custom providers.
- Steering handlers hit 100% accuracy in a 600-run benchmark versus 82.5% for prompts.
- Built-in hooks, OpenTelemetry tracing, MCP client, guardrails, and structured output.
- Deploys to AgentCore, Lambda, Fargate, EKS, or Docker with the same agent code.
- Production users include Smartsheet, Swisscom, Eightcap, Tavily, and Verisk Analytics.
Building AI agents usually starts simple and ends as a wall of prompt instructions and glue code. Strands Agents, an open source SDK originally built inside Amazon and now maintained publicly, takes the opposite bet: keep the loop tiny, let the model drive, and provide sharp tools for controlling it when it drifts.
The project ships as a monorepo with Python and TypeScript SDKs, documentation, and an MCP server. Strands takes a model-driven approach to building and running agents, from conversational assistants to autonomous workflows. It is Apache-2.0 licensed and has crossed roughly 7,000 GitHub stars.
The three-line agent
An agent in Strands consists of a model, a set of tools, and a loop. Nothing more is required to get started:
from strands import Agent
from strands_tools import calculator
agent = Agent(tools=[calculator])
agent("What is the square root of 1764")Both SDKs default to the Amazon Bedrock model provider, so you'll need AWS credentials configured and model access enabled for Claude Sonnet. Swapping backends does not change your code. First-class support covers Amazon Bedrock, Anthropic, OpenAI, and Gemini, plus many more providers and custom ones. MCP, streaming, multi-agent patterns, and structured output via Pydantic or Zod are built in.
This story is for Pro members
You've reached the end of the free preview. Upgrade to AlphaSignal Pro to read the full article - and everything else behind the paywall.
Comments
