Building and Operating AI Agents

• Traditional automation versus agentic AI
• Single prompts, fixed workflows, copilots, and agents
• Degrees of autonomy
• Selecting tasks suitable for agents
• Defining the problem and boundaries before building

Hands-on Lab: Workflow, Copilot, or Agent? — Classify realistic work tasks by required autonomy and determine which should use a prompt, fixed workflow, copilot, or AI agent.

• Defining goals and success conditions
• Agent instructions and operating boundaries
• Tools and connectors
• Memory and state
• Stop conditions
• Credentials and permissions
• Data-access restrictions
• Protection against hidden or malicious instructions
• AI-agent use cases in IT operations and service management

Hands-on Lab: Build Your First Agent — Create a working agent in a no-code agent platform, connect it to approved tools, and demonstrate successful completion of defined tasks.

• Browser-enabled and computer-using agents
• Tool execution across multiple systems
• Agent handoffs
• Multi-agent orchestration
• Assigning specialized roles to multiple agents
• Coordinating outputs between agents

• Sequential and parallel workflows
• Human-in-the-loop architectures
• Approval gates before external actions
• Escalation conditions
• Reviewing logs and agent actions
• Monitoring handoffs
• Cost and usage controls
• Failure recovery, retries, timeouts, and safe rollback

Hands-on Lab: Add Guardrails — Configure approval steps, operating limits, access restrictions, and other guardrails around the agent built in the previous lab.

• Building agent test cases
• Normal and edge-case behavior
• Missing or ambiguous data
• Failed tools and external-system errors
• Unsafe or unauthorized actions
• Regression testing after agent changes
• Optimizing behavior while preserving controls

Hands-on Lab: Stress-Test Your Agent — Run structured test scenarios against the agent, document failures, apply fixes, retest the system, and estimate operational usage cost.

• Building a proof of concept
• Deployment requirements
• Risk and compliance involvement
• User roles and responsibilities
• Implementation runbooks
• Technical and user readiness
• Controlled release practices

• Monitoring agent execution
• Operational metrics
• Scaling challenges
• Cost monitoring
• Measuring automation impact
• Measuring time to value
• Identifying agents that should be improved, restricted, or retired

Deliver a working AI agent together with its design specification, configured connectors, guardrails, test evidence, monitoring approach, operating instructions, and launch plan. The capstone is assessed against a skills-based rubric and must be successfully completed to earn the CERT4TECH Certificate in Building and Running AI Agents.