AI Agents for Workflows

AI agents are a new class of artificial intelligence tools that can plan, act, and adapt across multiple steps to complete tasks. As these tools become more common in workplace technology, organizations are increasingly exploring how they fit into real workflows. This course offers a practical, no-code introduction to agentic AI for professional settings. Participants will learn what AI agents are, how they differ from traditional automation and chatbots, how they work at a high level, and how to evaluate whether a tool is right for a specific workflow. 

The course also explores the current agentic AI landscape, permission models, security considerations, lifecycle management, and workforce readiness, with an emphasis on evaluation, oversight, and informed decision-making rather than coding or technical development.

What You Will Learn

• Explain what AI agents are and how they differ from automation, chatbots, and decision-support tools.
• Describe how agentic tools work at a high level, including the AI model, planning loop, tool access, and containment boundaries.
• Evaluate whether a specific agentic tool is appropriate for a given workflow using a structured framework.
• Identify risks, limitations, and operational warning signs associated with agentic AI systems.
• Apply human oversight models and governance frameworks to agentic tool use.
• Assess organizational readiness for deploying agentic tools.
• Recognize security considerations specific to agentic tools, including prompt injection and permission models.

Course Outline

Module 1: Understanding AI Agents

• What AI agents are and how they differ from automation, chatbots, and decision-support tools
• Core characteristics of AI agents: goal-driven behavior, multi-step execution, autonomy, and context-awareness
• The agentic AI landscape today: desktop agent tools, enterprise agent platforms, developer tools, and AI assistants with tool access
• How agentic tools work under the hood: the AI model, planning loop, tool access, and containment boundary
• Live demonstration of an agentic tool performing a multi-step government workflow task
• Human oversight roles: human-in-the-loop, human-on-the-loop, and human-in-command
• Common misconceptions and key risks, including prompt injection
• Current federal AI policy direction and its implications for government agencies

Module 2: Evaluating Agentic Tools for Government Work

• The evaluation mindset: understanding the workflow before evaluating the tool
• Identifying agent roles in a workflow: intake, analysis, recommendation, and escalation
• The Agent Evaluation Blueprint: goal, trigger, inputs, actions, boundaries, and oversight
• Permission models and data access: folder-level vs. organization-wide, network access, and least privilege
• Prompt injection: what it is, how it works, and why agentic tools are uniquely vulnerable
• Questions to ask before saying yes: a practical pre-approval checklist
• Hands-on activity: evaluate a realistic agentic tool proposal for a government workflow

Module 3: Where Agentic Tools Fit — and Where They Don't

• Appropriate use cases: case triage and routing, document analysis and extraction, internal coordination, and monitoring and alerts
• High-risk or inappropriate uses of agentic tools in government
• Operational risks: over-automation, over-reliance, and rubber-stamping
• Drift: data drift, concept drift, and objective drift — and how to detect them
• Early operational warning signs that an agentic tool may be failing
• Hands-on activity: red team a deployed agentic tool scenario to identify risks and recommend action

Module 4: Governing Agentic Tools in Your Organization

• Why governance is essential — and why no centralized federal AI regulator is coming
• Governance vs. technical controls: policies, oversight bodies, and accountability structures
• Legal, ethical, and procurement considerations: FedRAMP, ATO, vendor data handling, and records retention
• Security for agentic tools: access controls, prompt injection defenses, anomaly monitoring, and incident response
• Lifecycle management: design, pilot, deploy, monitor, update or retire — including regulatory sandbox alignment
• Performance monitoring and metrics: accuracy, override rates, equity indicators, and user satisfaction
• Workforce readiness and change management: role clarity, training on tool limitations, and avoiding fear and over-trust
• Deployment readiness checklist: a practical gate review before any agentic tool goes live
• Hands-on activity: conduct a readiness gate review for a proposed agentic tool deployment