# Microsoft Agent Framework technology due diligence

- **Agent pattern:** Discover-and-verify loop
- **Search surfaces:** News and web
- **Saved artifact:** `microsoft-agent-framework-due-diligence.md`


# What you'll build

A Microsoft Agent Framework analyst that discovers current warehouse-robotics developments, verifies material company claims, labels uncertainty, and turns the evidence into a technology due-diligence memo.

**Default brief:** Find five recent developments, verify the strongest company claims, separate verified facts from vendor claims and analyst inference, and finish with adoption risks.


# How the agent works

1.  **Discover developments.** News search gathers a focused set of recent warehouse-robotics signals.
2.  **Verify material claims.** Web search checks durable sources behind the most consequential company statements.
3.  **Label the evidence.** The agent separates facts, vendor claims, and inference before evaluating adoption risk.


# Core agent setup

This excerpt keeps Microsoft Agent Framework's client, instructions, and tools together. The full script also validates credentials and writes the resulting due-diligence memo.

``` python
agent = Agent(
    client=OpenAIChatClient(model=MODEL, api_key=api_key),
    name="technology_due_diligence_agent",
    instructions=(
        "Use live SerpApi tools, cross-check material claims, "
        "label uncertainty, and include direct source URLs."
    ),
    tools=[
        news_search(provider="microsoft-agent-framework"),
        web_search(
            provider="microsoft-agent-framework",
            allowed_engines=["google_light", "bing"],
        ),
    ],
)
report = str(await agent.run(PROMPT))
```

[View the complete `main.py` →](https://github.com/serpapi/serpapi-search-tools-python/blob/main/cookbook/microsoft-agent-framework/main.py)


# Run the recipe

Set `SERPAPI_API_KEY` and `OPENAI_API_KEY` in the repository-root `.env`, then run:

``` bash
uv run --isolated --no-project --with 'serpapi-search-tools[microsoft-agent-framework]' --with python-dotenv \
  cookbook/microsoft-agent-framework/main.py
```

Optional controls: `OPENAI_MODEL`, `COOKBOOK_PROMPT`, and `COOKBOOK_OUTPUT_DIR`.

**One agent, two evidence passes:** The framework controls tool use, while the instructions make discovery and verification separate responsibilities.


# Inspect the result

The recipe writes `cookbook-output/microsoft-agent-framework-due-diligence.md`. Expect five current developments, cross-checked claims, explicit evidence labels, adoption risks, and direct source URLs.

Use `COOKBOOK_PROMPT` to apply the same diligence pattern to another technology or market.


# SerpApi tools used

[news_search](../../reference/news_search.md#serpapi_search_tools.news_search) discovers recent warehouse-robotics developments, and [web_search](../../reference/web_search.md#serpapi_search_tools.web_search) checks durable sources behind the strongest company claims. This sequence helps the agent label verified facts, vendor statements, and analyst inference separately.

**Framework pattern:** The agent tool loop follows Microsoft Agent Framework's Python tool-use example and uses SerpApi news and web tools.

- [Open the runnable recipe](https://github.com/serpapi/serpapi-search-tools-python/tree/main/cookbook/microsoft-agent-framework)
- [View Microsoft's source example](https://github.com/microsoft/Agent-Framework-Samples/blob/main/00.ForBeginners/04-tool-use/code_samples/python-agent-framework-ghmodel-tools.ipynb)