Agentic AI Notes

Deterministic vs dynamic pipelines, and when a workflow beats an agent.

How an agent selects, validates and executes tools safely.

Short-term, long-term, semantic and episodic memory patterns.

State management, persistence, checkpoints and recovery.

Task decomposition, plan-and-execute, and reflection patterns.

Agentic retrieval — deciding when and how to search, not just retrieving once.

Supervisor patterns, handoffs, and when multiple agents beat one.

Routing, supervision and human-in-the-loop approval for high-impact actions.

Building stateful agent graphs — nodes, edges, checkpointing and memory.

The Model Context Protocol — what it standardizes, and what it doesn't.

LangGraph, LangChain agents and other frameworks compared conceptually.

Evaluating trajectories and tool use, not just the final answer.

Tracing agent trajectories, tool calls, cost and latency.

Prompt injection, least privilege, sandboxing and credential security for agents.

Preventing infinite loops, handling tool failures, retries and fallbacks.

Deploying, scaling, versioning and cost-optimizing agents in production.

Design patterns and engineering tradeoffs for agent systems.

Real-world agent roles — support, research, coding, sales and more.

Topic-wise Agentic AI interview questions with explained answers.

Implementation-oriented exercises across tools, memory, RAG and MCP.

End-to-end agent builds — research, support, SQL analyst, multi-agent systems.