For this walk-through, you just need Python installed and access to AgentCore with Bedrock. We're keeping it simple, no overcomplicated examples, just the core concepts you need to get started.
Setting up Strands is pretty simple. Let’s start with installing:
pip install strands-agents strands-tools
Once you get that installed, actually building and running an agent is about as simple as typing it. BUT there is a check list you must run through before your agent can do its thing.
Pre-flight checklist:
System prompts will need to be crafted in order for your agent to know what it can do. This is your agent's instruction manual. Be specific about what it should do and how it should behave.
SYSTEM_PROMPT = """You are a helpful assistant that can:
Make HTTP requests to external APIsProcess and format dataRespond conversationallyWhen using external APIs:
- Parse responses carefully
- Handle errors gracefully
- Format data in user-friendly ways
"""
"
Whether you use Strands or not, defining clear instructions for AI agents should be a habit if it's not already.
Tools extend your agent's capabilities beyond just text generation. The http_request tool lets your agent interact with any HTTP API on the web.
from strands import Agent
from strands_tools import http_request
agent = Agent(
system_prompt=SYSTEM_PROMPT,
tools=[http_request], # This enables HTTP capabilities
)
The magic of Strands is that users interact through natural language. They don't write API calls, they just ask questions.
response = agent("What's the current status of X?")
print(response)
The agent figures out it needs to make an API call, executes it, parses the response, and gives you a conversational answer.
When you send a message to your agent:
All of this happens automatically. You just ask questions like you're talking to a person.
Be specific in system prompts. Vague instructions like "you can look up data" won't cut it. Tell your agent exactly which APIs to use and how to format responses.
Start simple, then scale. Get one API working first. Then add more capabilities. Then handle edge cases.
Handle errors in the prompt. Tell your agent what to do when APIs fail, rate limits hit, or data is missing.
Test edge cases. Try queries that might fail. See how your agent handles unexpected responses.
Forgetting the tools parameter - If you don't pass tools=[http_request], your agent can't make HTTP calls.
Vague system prompts - "You can look up information" is too vague. Be specific about APIs, parameters, and expected behavior.
Not handling rate limits - Many APIs have usage restrictions. Tell your agent how to handle them.
Skipping error cases - APIs fail. Teach your agent to handle errors gracefully.
The real power of Strands comes from chaining operations. Your agent can:
The agent handles the orchestration. You just describe the workflow in your system prompt.
Remember, if you need to handle authentication, custom headers, or complex error handling, you have a good idea how to do so! Just update your system prompt with those instructions.
Here are ways to extend your agents:
Create your own tools - Premades like http_request are great, but custom tooling is the next step.
Connect multiple APIs - Your agent can query different services and combine their data
Build workflows - Chain multiple operations together for complex tasks
Add context - Use conversation history to make smarter decisions
Additional Resources: