Agentic AI Patterns.

Multi-agent architectures, tool use, orchestration workflows, memory systems, evaluation, and guardrails. Covers 11 sub-skills.

• Agentic AI Patterns · 1 of 3

  Tool Use & Function Calling

  What must the agent provide for a tool call?

  AAn expected output format together with a hard per-call timeout valueBThe tool name and arguments matching its expected input schemaCVerified user identity along with scoped authorization credentialsDThe full deployment environment plus the compiled source for the target

• Agentic AI Patterns · 2 of 3

  Model Context Protocol (MCP)

  How should MCP servers keep per-client state across tool calls?

  AReturn state inside each result for the client to pass back next callBEncode all state into each call's parameters so calls stay statelessCPersist state to a shared global database after every single callDSession-scoped memory keyed by connection ID, cleaned on disconnect

• Agentic AI Patterns · 3 of 3

  Model Context Protocol (MCP)

  How implement transactions across multiple MCP tool calls?

  ABatch the dependent calls into one atomic request applied togetherBSaga pattern with compensating actions for eventual consistencyCTreat transactions as unneeded since each call is atomic on its ownDRely on built-in begin/commit/rollback verbs the protocol defines