Designing Reliable AI Agents: Architecture, Control, and Evaluation
Advanced29 hours10 sections49 lessonsRated 4.7 out of 5 from 11 ratings
Design agents that use narrow tools, preserve durable state, stop on clear conditions, and act only within granted authority. The course covers retries, approvals, sandboxing, evaluation, incident stops, and the cases where a fixed workflow is the better design.
What you will be able to do
- Define an agent's observations, actions, state, and authority.
- Design tool permissions, approvals, retries, and clear stopping conditions.
- Evaluate agent behavior with evidence rather than demo success.
- Recognize when a fixed workflow is the better design than an agent.
Who this course is for
Engineers designing agentic systems, or interviewing on them, who need to reason about control, safety, and evaluation.
Helpful to know first
- Familiarity with LLM applications, prompting, and tool calling.
Course content
10 sections 49 lessons 29 hours
Define an agent’s observations, actions, state, and authority. Compare decision mechanisms and architectures, then examine tool permissions, evaluation, and operational controls.
- Introduction to AI Agents17 min
- Agent Architecture: Core Agent Components13 min
- Agent Architecture: Components Interaction and Agent Memory13 min
- Structuring Agent Behavior: Agent Orchestration Patterns16 min
- Building Trustworthy Agents: Guardrails and Human Oversight21 min
- Key Challenges and Design Strategies in Agentic AI Systems18 min
Locked sections open with Premium.Upgrade to Premium
Rate this course
Rated 4.7 out of 5 from 11 ratingsTell us how useful this course was. You can update your rating later.
Loading your rating...
Course rating
Rated 4.7 out of 5 from 11 ratings
- 5 stars: 8 ratings
- 4 stars: 3 ratings
- 3 stars: 0 ratings
- 2 stars: 0 ratings
- 1 star: 0 ratings