AI Free course · Certificate included
Building Production AI Agents
Design, build, secure, evaluate and ship AI agents that work reliably with real tools, data and people
- Advanced
- 8 h 26 min
- 18 lessons in 7 modules
- 2 h 42 min of video lectures
- Updated Sep 2026
About this course
Agents are moving from demos to production, and most fail for engineering reasons, not model reasons. This advanced course teaches you to build agents that work reliably. You will learn when an agent is the right choice and when a workflow is better, the core architectures (prompt chaining, routing, orchestrator–workers, evaluator–optimizer, multi-agent), and how to hand-build the agent loop in Python. You will design tools models use correctly, tune reasoning effort, engineer context and memory, and make long-running agents durable and idempotent. You will add human approvals, defend against prompt injection with least privilege and validators, evaluate outcomes and trajectories with pass^k, and instrument agents with OpenTelemetry. Finally you will cut cost and latency, deploy with sandboxes and staged rollouts, compare today's agent SDKs (Claude Agent SDK, OpenAI Agents SDK, Google ADK, LangGraph, CrewAI, Microsoft Agent Framework), and build and launch a research-and-ops agent end to end.
Tools you’ll use
- Claude API
- Claude Agent SDK
- OpenAI Agents SDK
- Google ADK
- LangGraph
- CrewAI
- Microsoft Agent Framework
- OpenTelemetry
- FastAPI
- Python
- SQLite
- MCP
Skills
- AI agent engineering
- Tool and function design
- Agent evaluation
- Prompt-injection defense
- LLM observability
- Context engineering
- Multi-agent systems
- LLM cost optimization
What you’ll be able to do
- Decide when to use an agent versus a workflow and choose the right architecture pattern
- Build a production-minded agent loop in Python with budgets, error handling and logging
- Design tools, plans and multi-agent delegation that models execute reliably
- Engineer context, memory, retrieval and durable state for long-running agents
- Implement risk-tiered approvals and layered defenses against prompt injection
- Evaluate agents on outcomes, trajectories, reliability (pass^k) and cost, with tracing
- Deploy agents with sandboxes, versioned config and staged rollouts, choosing the right SDK
Curriculum
Syllabus
- Modules
- 7
- Lessons
- 18
- Reading time
- 5 h
- Assessment questions
- 30
What agents are, the pattern catalog from prompt chaining to multi-agent systems, and a hand-built agent loop you fully understand.
- What an AI agent really is (and when not to build one)Video lecture, 9′14 min
- Workflow and agent architectures: the pattern catalogVideo lecture, 9′16 min
- Build the agent loop from scratch in PythonVideo lecture, 9′18 min
How to design tools agents use correctly, tune built-in reasoning and plans, and decide when multiple agents beat one.
- Designing tools agents use correctlyVideo lecture, 9′16 min
- Planning, reasoning models and effort controlVideo lecture, 9′15 min
- Multi-agent systems: orchestration, handoffs and delegationVideo lecture, 9′16 min
Context engineering, memory design, agentic retrieval with verified citations, and making long-running agents durable and idempotent.
- Context engineering and agent memoryVideo lecture, 9′16 min
- Agentic retrieval: giving agents knowledge they can trustVideo lecture, 9′15 min
- State, durability and long-running agentsVideo lecture, 9′15 min
Risk-tiered approvals, escalation design, prompt-injection defense, least privilege and red-teaming for agents that act on real systems.
- Human-in-the-loop: approvals, escalation and reviewVideo lecture, 9′15 min
- Guardrails, permissions and prompt-injection defenseVideo lecture, 9′17 min
Measuring agents on outcomes, trajectories, reliability and cost, and instrumenting them with traces, dashboards and alerts.
- Evaluating agents: task success, trajectories and judgesVideo lecture, 9′17 min
- Observability: tracing, logging and dashboards for agentsVideo lecture, 9′15 min
Cutting cost and latency, deploying agents as reliable services with sandboxes and safe rollouts, and choosing among today's agent SDKs and frameworks.
- Cost and latency engineering for agentsVideo lecture, 9′16 min
- Deploying agents: architectures, sandboxes and safe rolloutVideo lecture, 9′16 min
- The agent SDK and framework landscape in 2026Video lecture, 9′15 min
Build a research and operations agent end to end in Python, then evaluate, trace, harden, red-team and prepare it for a staged launch.
- Capstone part 1: build a research and ops agent end to endVideo lecture, 9′25 min
- Capstone part 2: evaluate, observe, harden and launchVideo lecture, 9′22 min
- Final assessment
Your certificate
Finish with a credential anyone can check
Earn the Certified Production AI Agent Engineer badge: The holder can design, build and ship production AI agents: choosing between workflows and agents, hand-building the agent loop, designing reliable tools, engineering memory and durable state, implementing human approvals and prompt-injection defenses, evaluating outcomes and trajectories with reliability metrics, instrumenting with tracing, controlling cost, and deploying with sandboxes and staged rollouts.
Completed all lessons and scored at least 80% on the final assessment.
- A public verification page
- A PDF certificate to download
- An Open Badge you can share
- One click to your LinkedIn profile
Final assessment
- 30questions drawn from a larger pool
- 45 mintime limit
- 80%pass mark
- 3attempts per 24 hours
Start learning today. It’s free.
Every lesson is free to read. A free account saves your progress, unlocks the final assessment and issues your certificate.