Code examples for the agentic AI patterns discussed in the excellent article Building Effective Agents by Erik Schluntz and Barry Zhang of Anthropic, inspired, ported and adapted from the code samples by the authors.
The examples rely on Pydantic AI, an agentic AI orchestration library that is particularly well-suited for implementing these patterns with clarity and precision.
These examples complement the original code accompanying the article and hopefully serve a dual purpose, demonstrating both the foundational agentic AI patterns discussed in the article and the use of Pydantic AI.
-
Copy
dot.env
to.env
-
Fill in the API keys for one or more of your LLM provider(s)
-
Set
AI_MODEL
to the model you want to use (see https://ai.pydantic.dev/models/ for supported providers and models)For example:
OPENAI_API_KEY=sk-123456789 AI_MODEL=openai:gpt-4o
or
GEMINI_API_KEY=ABC-123_xyz AI_MODEL=gemini-1.5-pro
or
ANTHROPIC_API_KEY=ABC-sk-123456789 AI_MODEL=claude-3-5-sonnet-latest
You can also omit the
AI_MODEL
variable, which will result in a random model from all the models you have API keys for being chosen (you can use that to play with different models and compare). -
You can optionally set the
LOGFIRE_TOKEN
variable to instrument Pydantic AI and follow the agent calls with Pydantic Logfire.