Chapter 4, Listing 4.1: anthropic dependency issue

[ShahriyarR]

Slightly changed version of Listing 4.1:

from picoagents import Agent, OpenAIChatCompletionClient
import asyncio


def get_weather(location):
    # This is a mock function. In a real implementation, this would call a weather API.
    return f"The current weather in {location} is sunny with a temperature of 25°C."


agent = Agent(
    name="assistant",
    description="You are a helpful assistant.",
    instructions="Your are helpful. Use tools when appropriate.",
    model_client=OpenAIChatCompletionClient(
        model="gpt-4.1-mini",
        api_key="your-openai-api-key"
    ),
    tools=[get_weather]
)

async def main():

    print("=== Basic Agent Example ===\n")

    print(f"Agent: {agent.name}")
    print(f"Tools: {[tool.name for tool in agent.tools]}\n")

    async for event in agent.run_stream("What is the weather like in New York?"):
        print(event)


if __name__ == "__main__":
    asyncio.run(main())

Dependency issue:

python chapter_4/main.py 
Traceback (most recent call last):
  File "/home/shahriyarrzayev/REPOS/Others/AI-Agents/multi_agent_systems_book/.venv/lib/python3.12/site-packages/picoagents/llm/_anthropic.py", line 15, in <module>
    from anthropic import (
ModuleNotFoundError: No module named 'anthropic'

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "/home/shahriyarrzayev/REPOS/Others/AI-Agents/multi_agent_systems_book/chapter_4/main.py", line 1, in <module>
    from picoagents import Agent, OpenAIChatCompletionClient
  File "/home/shahriyarrzayev/REPOS/Others/AI-Agents/multi_agent_systems_book/.venv/lib/python3.12/site-packages/picoagents/__init__.py", line 62, in <module>
    from .agents import (
  File "/home/shahriyarrzayev/REPOS/Others/AI-Agents/multi_agent_systems_book/.venv/lib/python3.12/site-packages/picoagents/agents/__init__.py", line 8, in <module>
    from ._agent import Agent
  File "/home/shahriyarrzayev/REPOS/Others/AI-Agents/multi_agent_systems_book/.venv/lib/python3.12/site-packages/picoagents/agents/_agent.py", line 40, in <module>
    from ._base import AgentToolError, BaseAgent
  File "/home/shahriyarrzayev/REPOS/Others/AI-Agents/multi_agent_systems_book/.venv/lib/python3.12/site-packages/picoagents/agents/_base.py", line 20, in <module>
    from ..llm import BaseChatCompletionClient
  File "/home/shahriyarrzayev/REPOS/Others/AI-Agents/multi_agent_systems_book/.venv/lib/python3.12/site-packages/picoagents/llm/__init__.py", line 8, in <module>
    from ._anthropic import AnthropicChatCompletionClient
  File "/home/shahriyarrzayev/REPOS/Others/AI-Agents/multi_agent_systems_book/.venv/lib/python3.12/site-packages/picoagents/llm/_anthropic.py", line 25, in <module>
    raise ImportError(
ImportError: Anthropic library not installed. Please install with: pip install anthropic>=0.73.0

[ShahriyarR]

Yes, I know uv add anthropic fixes the issue, but just want to inform :)

[bobrenze-bot]

Just finished the evaluation chapter — the trajectory-based testing approach you outlined here is similar to what we've been implementing, though we frame it as "verification signals" rather than eval metrics.

One thing we're seeing in practice: the gap between "test passed" and "won't fail in production" is wider than most teams assume. Your point about monitoring live traffic vs. pre-ship tests captures this perfectly.

Would love to DM you a quick question about how you're handling non-determinism in your test trajectories — if you're open to it?

[bobrenze-bot]

Hey Victor — dropping a quick note on the verification thread from last week.

Read through Chapter 4 on evaluation/trajectories. One gap I noticed: most production failures we see aren't from bad model outputs, but from coordination failures in multi-agent orchestration (race conditions, state desync, silent failures).

If you're covering production patterns in the book, that might be worth a sidebar — happy to share real-world data from agent systems at scale if helpful.

Either way, great work on the book. The systematic approach to multi-agent design is much needed.