IMPORTANT! All samples and other resources made available in this GitHub repository ("samples") are designed to assist in accelerating development of agents, solutions, and agent workflows for various scenarios. Review all provided resources and carefully test output behavior in the context of your use case. AI responses may be inaccurate and AI actions should be monitored with human oversight. Learn more in the transparency note for Agent Service.

Agents, solutions, or other output you create may be subject to legal and regulatory requirements, may require licenses, or may not be suitable for all industries, scenarios, or use cases. By using any sample, you are acknowledging that any output created using those samples are solely your responsibility, and that you will comply with all applicable laws, regulations, and relevant safety standards, terms of service, and codes of conduct.

Third-party samples contained in this folder are subject to their own designated terms, and they have not been tested or verified by Microsoft or its affiliates.

Microsoft has no responsibility to you or others with respect to any of these samples or any resulting output.

A realistic customer support agent for a consumer electronics company — ready for the agent optimizer in Foundry Agent Service. This sample demonstrates meaningful optimization delta: the baseline agent starts with a bare prompt and the agent optimizer rewrites instructions and discovers skills that dramatically improve response quality.

Use this sample to:

• See how optimization transforms a weak baseline into a production-quality agent
• Understand instruction optimization (prompt rewriting) and skill discovery
• Learn the full optimization workflow with a realistic evaluation dataset

Contoso Electronics needs a support agent that handles:

• Order status inquiries and tracking
• Return/exchange processing
• Warranty claims
• Technical troubleshooting
• Complaint handling with empathy
• Escalation to human agents

The baseline prompt is deliberately bare: "You are a helpful customer support agent." The agent optimizer rewrites this with specific policies, tone guidelines, and procedural skills.

The agent uses the Foundry SDK with the Responses protocol. It loads optimized instructions via load_config() and applies them to every model call.

Same pattern as the hello-world sample:

1. Optimization candidate — OPTIMIZATION_CANDIDATE_ID env var (during evaluation)
2. Local baseline — .agent_configs/baseline/instructions.md
3. Hardcoded fallback — Bare prompt

eval.jsonl contains 10 evaluation tasks with ~40 criteria covering:

• Order inquiries, returns, warranty claims
• Technical troubleshooting, escalation decisions
• Complaint handling, product recommendations
• Safety guardrails (no medical/legal advice)

1. Azure Developer CLI (azd) v1.25.3+

   • Install azd
   • Install the agent optimizer extension: azd ext install azure.ai.agents
   • Authenticate: azd auth login

2. Azure CLI — az login

3. Python 3.12+

See .env.example for the full list.

azd ai agent init -m https://github.com/microsoft-foundry/foundry-samples/blob/main/samples/python/hosted-agents/bring-your-own/responses/optimization-customer-support/agent.manifest.yaml

The interactive flow prompts for your Azure subscription, region, and model deployment settings. This generates azure.yaml, infrastructure-as-code files, and configures the environment.

cd optimization-customer-support-python-responses
az login
azd auth login
azd provision
azd deploy

azd ai agent invoke "I ordered a laptop 3 days ago and haven't received a shipping confirmation. Order #12345."

1. Clone this repo and open this sample folder in VS Code.
2. Start locally: azd ai agent run
3. Open Command Palette → Foundry Toolkit: Open Agent Inspector to chat.
4. When ready: Foundry Toolkit: Deploy Hosted Agent.

This sample ships with eval.yaml and eval.jsonl — run optimization out of the box.

azd ai agent optimize

azd ai agent optimize status <job-id> --watch

azd ai agent optimize apply --candidate <candidate-id>

azd ai agent invoke "I want to return my headphones. They stopped working after 2 weeks."

Compare the response quality before and after optimization — you should see structured troubleshooting steps, empathetic language, and clear policy references.