AI Agents (Agentic AI): Tool Use, MCP, Multi-Agent Systems and Evaluation

Learn agentic AI live: agent loops, tool use, the Model Context Protocol, memory, planning, LangGraph, CrewAI and AutoGen, human in the loop, evaluation, guardrails and deploying agents.

Live batches with a trainer · Level: Intermediate · Duration: 40 hours · Language: English · Fee: ₹24,999

About this course

AI agents are the next step after chatbots: systems that plan, call tools, read and write data and finish tasks on their own, with a person in the loop where it matters. In 2026 every company is piloting agents for support, operations, sales and IT, and engineers who can build them reliably are the scarcest AI profile.

The course starts with what an agent is (the loop of reasoning, tool calls and observation), the model APIs' tool use and structured output, and building a first agent in plain Python. You then learn the Model Context Protocol (MCP) to give agents tools and data in a standard way (servers, clients, resources, prompts), memory (short-term, long-term, vector), planning and reflection patterns, and the frameworks: LangGraph for stateful graphs, CrewAI and AutoGen for multi-agent teams, and the Claude Agent SDK.

You then make agents safe and measurable: human-in-the-loop approvals, permissions and sandboxing, guardrails, evaluation with trajectories and task suites, observability and cost, and deployment as services with queues and retries, with browser and computer-use agents and SAP Joule agents as examples. The course ends with two portfolio projects (a support agent and an operations agent) and preparation for AI engineer interviews. Every topic is built live, and every session is recorded.

What you will learn

  • Understand the agent loop: reasoning, tool calls, observation and stopping; when an agent beats a workflow
  • Use tool use and structured output in the Claude, OpenAI and Gemini APIs and build an agent in plain Python
  • Build and use Model Context Protocol (MCP) servers and clients for tools, resources and prompts
  • Give agents memory: conversation, long-term and vector memory; planning and reflection patterns
  • Build stateful agents with LangGraph and multi-agent teams with CrewAI and AutoGen
  • Add human-in-the-loop approvals, permissions, sandboxing and guardrails
  • Evaluate agents with task suites and trajectories, and add observability and cost control
  • Deploy agents as services with queues, retries and monitoring
  • Finish two portfolio projects and prepare for AI engineer interviews

Requirements

  • Python and basic experience with an LLM API (the Generative AI course covers this)
  • Best suited to developers, generative AI engineers, automation and integration professionals and data scientists
  • A laptop with a stable internet connection (8 GB RAM is enough; cloud notebooks are used for heavy work)

Syllabus

  • Module 1: What an agent is (4 hours)
  • Module 2: Tools and the Model Context Protocol (6 hours)
  • Module 3: Memory, planning and reflection (5 hours)
  • Module 4: Frameworks (8 hours)
  • Module 5: Safety, humans and guardrails (5 hours)
  • Module 6: Evaluation and observability (4 hours)
  • Module 7: Deployment and operations (4 hours)
  • Module 8: Projects and interviews (4 hours)

Upcoming live batches · Questions and answers · Book a free demo