Helpful foundation
- Basic Python variables and functions
- Simple JSON and API concepts
- Linux/terminal familiarity
Agentic AI Engineering · AAE-FND
From agent basics to controlled production systems.
Learn Agentic AI from first principles: agents versus workflows, tool calling, state and memory, planning, multi-agent patterns, human approval, reliability, MCP, security, observability, evaluation, failure recovery and production architecture. Finish with a controlled Operations Investigation Agent capstone.
My Learning progress
0% complete
0 of 16 curriculum modules completed · 0/16 currently published modules complete · 0 with lab + knowledge-check evidence.
Beginner-first learning contract
Understand the control flow before the framework.
Every module moves from a plain-language problem to a mental model, worked examples, an account-free lab, a deliberate failure/boundary test and the production lesson.
WHY→PLAIN LANGUAGE→VISUAL MODEL→EXAMPLE→TRY IT→BREAK / FIX→PRODUCTION VIEW
Who can start?
No agent-framework or MCP experience is required.
Basic Python helps with the labs. All 16 core labs are account-free and provider-neutral; no hosted AI subscription or GPU is required.
You do not need first
- LangGraph/CrewAI/AutoGen expertise
- MCP SDK experience
- Vector databases
- Cloud AI subscriptions
Architectural rule
If the exact steps are known, prefer a workflow. Introduce agentic control only where bounded dynamic decisions create real value.
Learner outcomes
Build the system, controls and evidence—not just a model call.
Choose deliberately between deterministic workflows, bounded agents and hybrid control flow.
Build explicit agent loops, narrow tools, task state and observable plans without relying on hidden reasoning traces.
Coordinate specialist agents while keeping responsibility, context and tool authority scoped.
Gate consequential actions with exact, auditable human approval.
Bound failures with deadlines, retries, budgets, idempotency and recovery checkpoints.
Use MCP as an interoperability boundary without confusing discovery with authorization.
Threat-model prompt injection, over-broad tools, secret exposure and confused-deputy behavior.
Instrument agent systems for task outcome, latency, retries, tool failures, approvals and audit evidence.
Evaluate task success, tool correctness, policy compliance, cost and recovery with regression gates.
Design and defend a production architecture with rollout, rollback, kill switches and a controlled operations capstone.
Complete curriculum
16 modules · foundations → production capstone
All modules are fully navigable and progress-tracked. The course deliberately progresses from simple control-flow decisions to production security, evidence and operations.
01020304050607080910111213141516
What Is an AI Agent?
Separate language models, chatbots, deterministic workflows and agents before choosing an architecture.
35–45 min · AvailableInside an Agent
Understand the small set of parts that produce agentic behaviour before learning any framework.
45–55 min · AvailableGiving Agents Tools
Learn how an agent reaches outside the model without turning model output into unrestricted execution.
50–60 min · AvailableState & Memory
Keep the right information across steps without calling every stored fact “memory”.
50–60 min · AvailablePlanning & Execution
Turn a broad goal into visible, bounded steps, execute one step at a time, and update the plan when evidence changes the situation.
55–65 min · AvailableWorkflows vs Agents
Choose deterministic control where the steps are known, introduce agentic decisions only where uncertainty requires them, and combine both in hybrid systems.
50–60 min · AvailableMultiple Agents
Use multiple agents only when responsibilities, context or permissions genuinely need separation; otherwise keep the system simpler.
60–70 min · AvailableHuman Approval
Pause consequential actions, show a human exactly what will happen, bind approval to that exact request, and revalidate before execution.
55–65 min · AvailableReliability Engineering
Make agent runs stop predictably, retry only the right failures, avoid duplicate side effects, and escalate when recovery is unsafe.
65–75 min · AvailableModel Context Protocol (MCP)
Understand what MCP standardizes, how the 2026-07-28 stateless protocol works at a high level, and why protocol connectivity does not replace authorization or agent safety.
70–80 min · AvailableAgent Security
Protect the agent boundary from untrusted instructions, over-broad tools, secret leakage and confused-deputy behavior.
80–90 min · AvailableObservability
Make agent decisions, tool activity, failures, latency and cost visible without logging hidden reasoning or sensitive payloads.
70–80 min · AvailableEvaluation
Test the whole agent system with representative scenarios instead of judging a few fluent answers by eye.
75–90 min · AvailableFailure Recovery
Recover from partial execution without blindly repeating side effects or losing the evidence needed to continue safely.
80–90 min · AvailableProduction Architecture
Assemble models, state, tools, policy, MCP, approvals, telemetry, evaluation and recovery into one production-shaped architecture.
90–105 min · AvailableOperations Agent Capstone
Combine investigation, tools, state, planning, approval, security, observability, evaluation and recovery in one end-to-end production-shaped scenario.
2.5–3.5 hours · AvailableLearning modules
FREE
All 16 modules are open learning content.
Hands-on labs
FREE
16 account-free labs including security, evaluation, recovery, architecture and the operations capstone.
Assessment
FREE
30 randomized questions, 35-minute server timer, pass ≥75%, maximum 2 attempts per 24 hours.
Course completion certificate
FREE
After completing all modules, core labs and knowledge checks, request a publicly verifiable completion credential. It does not represent an assessment score.
Request completion certificateAssessment-verified certificate
OPTIONAL · ₹499 incl. GST
Certification follows a qualifying timed assessment. It records GNU Group Learning Academy assessment completion; it is not vendor accreditation or a professional licence.
Knowledge assessment
Apply the control model.
The assessment uses practical Agentic AI situations. Focus on control flow, evidence, safe tool use, authorization, approvals, reliability, security, evaluation and production operations.
The 84-question bank spans seven domains and contains 42 scenario/application questions. Each 30-question attempt preserves domain quotas and guarantees at least 14 scenarios.
16 MODULES→30 QUESTIONS · ≥14 SCENARIOS→35 MIN · PASS ≥75%→2 ATTEMPTS / 24H→OPTIONAL CERTIFICATION
2.5–3.5 hour production capstone
Investigate, propose, approve, execute once, verify and report.
The Operations Agent capstone combines read-only evidence gathering, bounded planning, exact approval, idempotent change, post-change verification and an auditable incident record. The capstone is formative; assessment-verified certificate eligibility is based on the timed Academy assessment.
ALERT→READ-ONLY EVIDENCE→PROPOSAL→POLICY + APPROVAL→EXECUTE ONCE→VERIFY→INCIDENT EVIDENCE