What you’ll learn
- Choose between deterministic workflows and model-directed steps.
- Implement bounded tools and state transitions.
- Evaluate task outcomes, failure recovery and review requirements.
Proposed course content
6 sections · 13 illustrative lesson records
Workflow design3 lessons
- Introduction to Agentic AI Application Engineering
Design agent workflows with explicit state, bounded tool access, evaluation and human review instead of unrestricted autonomous actions.
Public introduction · Video not supplied - Define task scope and observable success criteria
Define task scope and observable success criteria. Module: Workflow design.
Enroll to access · Video not supplied - Compare deterministic routing and agent decisions
Compare deterministic routing and agent decisions. Module: Workflow design.
Enroll to access · Video not supplied
State and context2 lessons
- Represent conversation and task state explicitly
Represent conversation and task state explicitly. Module: State and context.
Enroll to access · Video not supplied - Limit retained context and protect sensitive inputs
Limit retained context and protect sensitive inputs. Module: State and context.
Enroll to access · Video not supplied
Tool contracts2 lessons
- Validate arguments and bound tool permissions
Validate arguments and bound tool permissions. Module: Tool contracts.
Enroll to access · Video not supplied - Require review before consequential side effects
Require review before consequential side effects. Module: Tool contracts.
Enroll to access · Video not supplied
Planning and recovery2 lessons
- Set step budgets, timeouts and stop conditions
Set step budgets, timeouts and stop conditions. Module: Planning and recovery.
Enroll to access · Video not supplied - Recover from failed tools without repeating actions
Recover from failed tools without repeating actions. Module: Planning and recovery.
Enroll to access · Video not supplied
Evaluation2 lessons
- Build synthetic tasks and baseline comparisons
Build synthetic tasks and baseline comparisons. Module: Evaluation.
Enroll to access · Video not supplied - Measure completion, error rates and resource use
Measure completion, error rates and resource use. Module: Evaluation.
Enroll to access · Video not supplied
Operational review2 lessons
- Log safe traces and inspect uncertain outcomes
Log safe traces and inspect uncertain outcomes. Module: Operational review.
Enroll to access · Video not supplied - Document handoff, rollback and limitations
Document handoff, rollback and limitations. Module: Operational review.
Enroll to access · Video not supplied
Requirements
- Python or TypeScript programming and basic API integration.
- Access to a permitted local model or model API; synthetic task data.
Tools and resources
- A development computer with resources appropriate to the proposed local exercises.
- Python or TypeScript, a permitted model interface and synthetic documents.
- A local tool service, test runner and safe tracing utilities; hosted model resources are not included.
Description
Design agent workflows with explicit state, bounded tool access, evaluation and human review instead of unrestricted autonomous actions.
Who this course is for
Application developers exploring bounded model-assisted workflows.
Example practical project
Proposed capstone: a local document-triage agent that uses read-only tools and queues a proposed classification for human approval.