Frequently Asked Questions

iLoveAgents is a content-first blog by Christian Glessner with practical patterns and playbooks for building production-grade AI agents on Microsoft Agent Framework and Microsoft Foundry—Python/.NET tutorials, multi-agent workflows, tools & grounding, and deployment guides

No. iLoveAgents is an independent community and learning effort. Opinions are the author’s and contributors’ own.

Absolutely. Open an issue or pull request on the repo, or reach out on LinkedIn—especially with real-world agent scenarios, telemetry lessons, or responsible-AI guardrail patterns.

Yes. Upcoming posts cover moving from local dev to managed Microsoft Foundry hosting, observability, cost controls, and human-in-the-loop patterns.

An open-source SDK and runtime for building AI agents and multi-agent workflows in Python and .NET. It unifies learnings from Semantic Kernel and AutoGen and integrates with Microsoft Foundry for deployment, observability, and governance.

Semantic Kernel emphasized connectors, memory, and reliability. The Microsoft Agent Framework brings those enterprise foundations together with first-class agent orchestration and a cleaner abstraction surface. Keep stable SK apps; start new agent projects on the Agent Framework.

The Agent Framework is model-agnostic. It ships chat clients for Azure OpenAI and the OpenAI API and can target any inference service that implements the Microsoft.Extensions.AI `IChatClient` interface (including local/OSS via adapters like Ollama). In Microsoft Foundry you can pick from the model catalog (Azure OpenAI plus other providers). For providers not in your catalog/region (e.g., Anthropic Claude Sonnet), call them via a compatible client or a small adapter.

Register tools via OpenAPI or use supported/partner connectors so agents can call external systems safely. For enterprise knowledge, ground agents with Azure AI Search or other sources surfaced through Foundry. Interop patterns like MCP and agent-to-agent (A2A) are supported across the ecosystem.

Public preview. Core concepts (agents, tools, messages, workflows) are converging, but expect some changes while the surface area settles. Pin package versions, watch the GitHub repo, and budget time for minor updates.

No rush. Keep stable SK catergoryion workloads. For new multi-agent or orchestration-heavy projects, use the Agent Framework. If migrating, start with a thin vertical slice, validate tool/memory parity, then roll forward incrementally.

Microsoft's unified platform to build, evaluate, secure, and operate generative-AI and agentic apps. It centralizes models (Azure OpenAI and catalog), orchestration, data/grounding, safety, evaluations, hosted endpoints, monitoring, governance, and lifecycle management in one workspace.

Design locally, then publish to Agent Service for managed hosting with identity, networking, secrets, safety filters, and observability. Use environments for config, run evaluations before promotion, and monitor telemetry and cost in catergoryion.

Yes. Foundry provides observability to inspect conversations, tool calls, token/latency metrics, and failures. You can export traces (e.g., OpenTelemetry) and iterate locally before promoting to catergoryion.

The framework is open source; you pay for model usage and Azure services (Agent Service hosting, storage, logging). Model pricing/limits/regions vary by provider. Foundry centralizes identity, networking, safety, and logging to support enterprise requirements.