Agentic AI From Fundamentals to Production

via Coursera Specialization

Go to class Write review

Overview

Google, IBM & Meta Certificates – 40% Off

One Coursera Plus subscription covers most Professional Certificates on Coursera.

Unlock All Certificates

"I think the course is excellent." — Dr. Peter Norvig, Researcher at Recursive / Education Fellow at Stanford & co-author of Artificial Intelligence: A Modern Approach "For those looking for an insightful, comprehensive outline of Agentic AI, this course is a must." - Dr. Anoop Sinha, Research Director, Google Agentic AI: From Fundamentals to Production takes you from core LLM concepts to designing, building, and operating autonomous AI agents across six courses and 20 modules. Start with LLM fundamentals, inference, and prompt engineering. Build agents from the ground up: architectures, tool use, MCP integration, memory, and RAG. Advance to planning, design patterns, frameworks, and multi-agent orchestration. Build responsibly with safety, privacy, and governance guardrails. Rigorously evaluate, benchmark, and test agent behavior before it ships. Deploy, optimize, and operate agents at scale with AI Ops. By the end, you'll be able to build reliable, cost-efficient agentic systems ready for real-world use.

Syllabus

Course 1: Course 1: AI, LLM & Prompt Engineering
Course 2: Course 2: Basic Agentic AI Concepts
Course 3: Course 3: Advanced Agentic AI
Course 4: Course 4: Responsible AI, Safety, & Governance
Course 5: Course 5: Agent Evaluation & Testing
Course 6: Course 6: Deployment, Optimization, & AI Ops

Courses

0 reviews

View details

Course 6: Deploying and Operating Agentic AI in Production ---------------------------------------------------------- What you will learn ------------------- (a) Deploy an agent as a reliable, scalable service (b) Cut cost and latency with caching, model routing, output limits, and batching (c) Monitor agents in production with dashboards, alerts, and drift detection (d) Manage the lifecycle: versioned prompts and models, rollbacks, and provider changes Why take it ----------- (a) Takes an agent the last mile, from notebook to monitored service with a cost budget (b) Every optimization is measured against its effect on quality What makes it unique -------------------- (a) Capstone of the specialization: the labs pick up the investment assistant built in earlier courses (b) Ends with a production checklist and an operating model you can adapt to your organization
0 reviews

View details

Course 4: Responsible Agentic AI: Ethics, Safety, and Governance ---------------------------------------------------------------- What you will learn ------------------- (a) Make agent decisions explainable with faithful reasoning and honest confidence (b) Defend against prompt injection and unsafe tool sequences (c) Protect personal data in agent memory and logs (d) Produce tamper-evident audit trails and a governance model with clear ownership Why take it ----------- (a) Agents act on the world, so the stakes go beyond a wrong chatbot answer (b) You attack your own agent first, then add the guardrails that break the chain What makes it unique -------------------- (a) Most labs are deterministic and run without a model API key, so every result is reproducible (b) Ends with a pre-release checklist you can apply to any agent
0 reviews

View details

Course 2: Basic Agentic AI Concepts ----------------------------------- What you will learn ------------------- (a) Design an agent from its core components (b) Connect tools through the Model Context Protocol (MCP) (c) Add short-term and long-term memory that persists across sessions (d) Ground answers in your own documents with RAG, reranking, and citations (e) Engineer the context window so the right information arrives at the right time Why take it ----------- (a) Turns a chat model into a working agent, one layer at a time (b) Teaches you which layer is broken when an agent misbehaves What makes it unique -------------------- (a) One running example, an investment assistant, built across all five modules (b) Labs run against real APIs and a real vector database, so you see failures before fixes
0 reviews

View details

Course 5: Agent Observability, Evaluation, and Quality Engineering ------------------------------------------------------------------ What you will learn ------------------- (a) Trace every agent step with OpenTelemetry-compatible tooling (b) Evaluate retrieval and generation separately with DeepEval and RAGAS (c) Write behavioral contracts and generate synthetic test sets from them (d) Use LLM-as-judge and audit it against human raters with Cohen's kappa (e) Gate releases on quality thresholds in CI Why take it ----------- (a) Agents fail silently and intermittently; instinct is not enough (b) Gives you a reusable evaluation harness and shared vocabulary with product and compliance teams What makes it unique -------------------- (a) Treats agent quality as engineering, with CI gates rather than vibe checks (b) Includes a live drift scenario: a defect is seeded, an alert fires, and you walk the trace to the root cause
0 reviews

View details

Course 1: AI & LLM Core Concepts and Prompt Engineering ------------------------------------------------------- What you will learn ------------------- (a) Explain how LLMs work, from transformers to the inference pipeline (b) Classify AI systems by capability and by the five levels of autonomy (c) Choose a model by architecture, tier, capability, and pinned version (d) Control output with temperature, top-P, penalties, and reasoning effort (e) Write prompts that produce reliable, structured results Why take it ----------- (a) Builds the foundation every later course assumes (b) Explains why identical prompts still give different answers, and what to do about it What makes it unique -------------------- (a) Provider-agnostic labs: call Anthropic, OpenAI, and Google models side by side on the same task (b) No prior AI experience required; reading Python is enough
0 reviews

View details

Course 3: Advanced Agentic AI ----------------------------- What you will learn ------------------- (a) Choose an agent framework with clear criteria (LangGraph, CrewAI, and others) (b) Build planning loops that monitor, replan, and abstain when evidence is missing (c) Apply ReAct, Algorithm of Thoughts, and evaluator-optimizer to specific failure modes (d) Coordinate multiple agents with supervisor, maker-checker, and peer protocols (e) Design interfaces and approval flows for human-agent collaboration Why take it ----------- (a) Moves you from single agent loops to multi-agent systems (b) Every pattern is measured for its cost in API calls, latency, and reliability What makes it unique -------------------- (a) Patterns are matched to the failure they fix, not presented as a catalog (b) Final module is optional for non-developers