AI Agents
Agentic AI
Create agents that reason over context, call tools, use MCP, and collaborate in controlled workflows.
Start with the first guideYour outcome
Deploy a useful agent with tools, evaluation, and guardrails.
Core curriculum
The skills to practice
Guided path
Agentic AI developer
Progress from a single tool-calling agent to MCP integrations and governed multi-agent workflows.
Suggested pace: 6 weeks · Capstone: Multi-agent operations desk
0 of 5 complete
- 01
Connect agents to tools deliberately
Choose a narrow tool boundary and record what data may cross it.
- 02
Design state and memory
Separate short-lived context, workflow state, preferences, and knowledge.
- 03
Add layered guardrails
Define permissions, approval points, adversarial tests, and recovery.
- 04
Choose orchestration only where needed
Compare a single-agent baseline with one justified specialist pattern.
- 05
Evaluate before expanding autonomy
Measure task completion, tool behavior, safety, latency, and cost.
Practical guides
Learn through decisions and builds.
MCP for Azure AI agents: when and how to use it
Understand Model Context Protocol, its role in agent tooling, and the decisions that matter in Azure deployments.
How to evaluate AI agents before production
Create an evidence-based evaluation loop for tool-using agents with quality, safety, latency, and cost measures.
Five multi-agent architecture patterns that hold up
Compare sequential, concurrent, handoff, supervisor, and group collaboration patterns with practical trade-offs.
Guardrails for enterprise AI agents
Design layered safeguards for agent identity, untrusted retrieved data, tool permissions, consequential actions, monitoring, and recovery.
Design memory for AI agents without creating risk
Design agent memory by separating conversation context, workflow state, user preferences, and external knowledge, with explicit access, retention, deletion, and evaluation rules.