AI Basics for Business Owners — Lesson 4
Chatbots, AI Agents, and Workflows
Learning Objectives
- 1Distinguish between chatbots, copilots, and AI agents.
- 2Understand when AI should assist versus act autonomously.
- 3Evaluate supervision requirements for different AI application types.
Chatbots: conversational interfaces
A chatbot is an AI system that interacts through conversation. Basic chatbots follow scripts and decision trees. AI-powered chatbots use language models to understand natural language and generate contextual responses. Customer support chatbots, FAQ assistants, and booking bots are common examples.
Chatbot quality varies enormously. A well-designed chatbot with access to your knowledge base, clear boundaries, and escalation to humans for complex issues can handle 60-80% of routine inquiries. A poorly designed chatbot that gives wrong answers or cannot escalate frustrates users and damages trust.
Before deploying a chatbot, define what it should handle, what it should escalate to humans, what tone it should use, what information it should never share, and how you will monitor its performance. An unmonitored chatbot can give incorrect answers at scale.
AI agents: autonomy with boundaries
An AI agent is a system that can take actions — not just generate text — based on goals you define. An agent might research a topic, book a meeting, process a refund, update a database, send an email, or execute a multi-step workflow. The key difference from chatbots is that agents do things in the real world, not just talk about them.
Agent capabilities are advancing rapidly, but so are the risks. An agent that can send emails on your behalf can also send embarrassing or harmful emails. An agent that can update your database can also corrupt it. An agent that can process refunds can also approve fraudulent ones.
The supervision principle: the more consequential the action, the more human oversight the agent needs. Low-consequence actions (scheduling a meeting, summarizing a document) can be more autonomous. High-consequence actions (sending money, communicating with customers, modifying production data) should require human approval.
AI in workflow design
The most practical AI applications are not standalone chatbots or autonomous agents. They are AI capabilities integrated into existing workflows at specific steps. AI generates a draft, a human reviews it. AI classifies a ticket, a human confirms the classification. AI suggests a response, a human personalizes and sends it.
This human-in-the-loop approach captures most of the time savings of AI while maintaining quality control. It is also the safest way to adopt AI because humans catch errors before they reach customers or affect business operations.
When planning AI integration into workflows, identify the specific step where AI adds value, define what the AI produces (draft, classification, recommendation, summary), define who reviews the output, and define what happens when the AI output is wrong or uncertain.
Case Study
The chatbot that went off-script
Situation
An airline deployed an AI chatbot for customer service. A frustrated customer asked about the refund policy and the chatbot, attempting to be helpful, made up a refund policy that did not exist and promised the customer a full refund they were not entitled to. The customer screenshot the conversation and posted it on social media, forcing the airline to honor the false promise.
Analysis
The chatbot had no guardrails limiting what policies it could communicate. It was designed to be helpful without being constrained to verified information. A system that limited responses to approved policy documents and escalated unusual requests to humans would have prevented this.
Takeaway
AI chatbots must be constrained to verified information and have clear escalation paths. An unconstrained chatbot that invents answers is a liability, not an asset.
Reflection Questions
- 1. If you deployed an AI chatbot for your business, what should it handle and what should it always escalate to a human?
- 2. For the AI tools your team uses, who reviews the output before it is sent to customers or used for decisions?
Key Takeaways
- ✓Chatbots converse; agents act. More autonomy requires more supervision.
- ✓Human-in-the-loop workflows capture AI speed while maintaining quality control.
- ✓Define what AI should handle, what it should escalate, and what it should never do.
- ✓An unmonitored AI system can create problems faster than it solves them.