Skip to Content
Home └AI Agents and Agentic Automation
AI Agents and Agentic Automation

AI Agents and Agentic Automation

Building an agent that works once in a demo takes an afternoon. Building one that runs unattended for three days across multiple departments, recovers from its own errors, and leaves an audit trail is an engineering discipline. This course is about the second thing.

Framework choice is taught as a decision rather than a preference. Five dimensions predict whether an agentic system survives production — cost, latency, efficacy, assurance and reliability — and delegates learn to evaluate candidates against those rather than against feature lists. The course covers LangGraph, CrewAI, Microsoft Agent Framework, OpenAI Agents SDK and Claude Agent SDK, with honest treatment of where each is weak.

MCP is taught early and used throughout, because it is the closest thing to insurance against lock-in. Tool integrations built against MCP port between frameworks almost trivially, so a team can swap orchestration layers without rewriting its integration surface.

Evaluation gets a full module. Most agent projects fail not because the agent cannot do the task but because nobody could measure whether it did the task correctly, so it never left pilot.

Self-paced course
₹15,000
or start with ₹5,000 and pay the balance within 3 months
₹
Pay in full ₹15,000
  • ○ Lessons, videos and quizzes from ₹5,000
  • 🔒 Lab recordings at ₹15,000
  • 🔒 Certification exam at full payment
  • ○ Optional real lab: ₹5,000 per month (after full payment)
Valid 12 months from your first payment. Refund policy
Responsible XertApp
Last Update 10/11/2026
Completion Time 1 day 21 hours
Members 1
Career Courses Hot Skills AI & Machine Learning
  • Module 1 — Agentic foundations (empty)
    5Lessons · 3 hr 45 min
    • Agents, assistants, workflows and automations distinguished
    • The reason-act loop and tool use
    • Where agentic approaches beat deterministic automation, and where they do not
    • Agentic automation against conventional RPA
    • Cost, latency, efficacy, assurance and reliability as design constraints
  • Module 2 — Working with language models (empty)
    5Lessons · 3 hr 45 min
    • Model selection: capability, cost and latency trade-offs
    • Context windows, token economics and prompt caching
    • Structured output and schema validation
    • Function and tool calling mechanics
    • Failure modes: hallucination, loop traps, refusal cascades
  • Module 3 — Model Context Protocol (empty)
    5Lessons · 3 hr 45 min
    • MCP architecture: servers, clients and transports
    • Consuming existing MCP servers for filesystems, databases and APIs
    • Building a custom MCP server
    • Authentication and credential handling
    • Portability benefits and lock-in avoidance
  • Module 4 — Single-agent construction (empty)
    5Lessons · 3 hr 45 min
    • Agent goals, instructions and scope
    • Tool design and description quality
    • Memory: short-term context and long-term stores
    • Retrieval-augmented generation for grounding
    • Error handling and retry semantics
  • Module 5 — Orchestration frameworks (empty)
    6Lessons · 4 hr 30 min
    • LangGraph: graph-based state machines, checkpoints and durable execution
    • CrewAI: role-based crews and rapid multi-agent assembly
    • Microsoft Agent Framework: the merged Semantic Kernel and AutoGen successor
    • OpenAI Agents SDK and Claude Agent SDK
    • LlamaIndex Workflows for document-heavy pipelines
    • Choosing a framework against real constraints
  • Module 6 — Multi-agent systems (empty)
    5Lessons · 3 hr 45 min
    • Coordination models: supervisor, hierarchy, peer network
    • Subagent spawning and delegation
    • Agent-to-Agent protocol and cross-organisation delegation
    • Shared state and handoff design
    • When multi-agent architectures make things worse
  • Module 7 — State, memory and long-running work (empty)
    5Lessons · 3 hr 45 min
    • Durable state and checkpoint APIs
    • Resuming workflows that span days
    • Rollback points and error correction loops
    • Conversation and session persistence
    • Human-in-the-loop approval gates
  • Module 8 — Evaluation (empty)
    5Lessons · 3 hr 45 min
    • Building evaluation datasets from real tasks
    • Task completion, trajectory and output quality metrics
    • LLM-as-judge and its limitations
    • Regression testing agents across model versions
    • Benchmarks and what they do not tell you
  • Module 9 — Safety, security and governance (empty)
    6Lessons · 4 hr 30 min
    • Sandboxed execution and least-privilege tool access
    • Prompt injection and untrusted content handling
    • Approval gates for consequential actions
    • Audit trails and explainability
    • Data residency and confidentiality constraints
    • Responsible AI obligations in an enterprise context
  • Module 10 — Production operations (empty)
    6Lessons · 4 hr 30 min
    • Deployment patterns and infrastructure
    • Observability, tracing and conversation review
    • Cost monitoring and token budget control
    • Latency optimisation and caching
    • Incident response when an agent misbehaves
    • Iterating on agents after release
  • Module 11 — Enterprise integration (empty)
    5Lessons · 3 hr 45 min
    • Connecting agents to ERP, CRM and ticketing systems
    • API integration patterns: REST, function calling, MCP gateways
    • Low-code orchestration alongside custom agents
    • Identifying and scoping candidate use cases
    • Building the business case and measuring return
  • Capstone (empty)
    2Lessons · 1 hr 30 min
    • Design, build, evaluate and deploy a multi-step agent against a real workflow
    • Written architecture justification and evaluation results

Who should attend

Aimed at engineers and technical leads who will build agentic systems, not only commission them.

  • Software engineers and backend developers building AI features
  • Data engineers and ML practitioners moving into agentic work
  • Integration and automation specialists moving beyond RPA
  • Solution architects designing AI-enabled systems
  • Technical leads evaluating frameworks and vendors
  • Consultants building agent capability for clients

Assumed knowledge: competent Python, familiarity with REST APIs and JSON, and comfort with version control and the command line. No prior machine learning background is required — this is a software engineering course rather than a data science one. Delegates without Python should take the AppSolWorld Python course first.

Delivery

Duration
40 hours, delivered as weekday evenings or weekend sessions.

Format
Live instructor-led online, with recordings available to delegates for the duration of the programme. On-site corporate delivery available on request.

Lab environment
Each delegate receives a development environment with model API access for the duration of the course and for four weeks afterwards.

Included
Course notes, lab guides, exercise solutions and a certificate of completion.

How to enroll

Enrol and pay online on this page. The full fee is ₹15,000 (US$170). You can start with ₹5,000 (US$60) and pay the balance in instalments within 3 months. Pay by card, UPI or net banking, or by PayPal in US dollars. Go to payment. For corporate, on-site or one-to-one training, contact contact@appsolworld.com or call +91 92051 38405.