AI Agent Workshop
The Big Idea
AI Agents are dynamic reasoning systems. They read, think, and respond based on patterns — which makes them remarkably flexible but inherently probabilistic. They can give you a brilliant answer, but they cannot guarantee it will be the same answer every time.
Businesses often need the opposite: 100% determinism. When a customer asks about their fees or their refund eligibility, the answer must be exact, consistent, and traceable — every single time. But not everything demands that level of rigidity. A product recommendation can afford to be conversational; a compliance-sensitive policy answer cannot.
This workshop puts that decision in your hands. Using a fictional European bank, you will build an AI Agent and see — live — how the way you structure and tag your knowledge articles directly controls whether the agent reasons freely, follows strict rules, or hands off to a person.
What You Will Experience
1. Narrowing what the agent knows, based on who is asking
Not every customer should see every piece of content. A basic-tier customer asking about premium services should not receive information about something they cannot access — because the agent would be giving an answer that is technically correct but practically wrong. In the workshop, you will see how a simple filter — chosen by the customer before the conversation starts — narrows the knowledge base so the agent only sees relevant articles. Same question, different customer profile, completely different answers. The filter is driven by a single tag that the knowledge author sets. No code, no AI configuration — just metadata.
2. Choosing where to be on the deterministic–dynamic spectrum
| Mode | How It Works | When to Use It |
|---|---|---|
| Fully Dynamic | The agent reasons freely across all available knowledge and tools. It interprets the question and crafts a natural response. | Open-ended questions: product recommendations, how-to guidance, general information. |
| Assured | AI Agent will confirm before proceeding with generating an answer. | Policy questions, fee schedules, eligibility rules — anywhere the answer must be traceable to an approved source. |
| Fully Deterministic | The agent follows a fixed, step-by-step process with no improvisation. Every step is pre-defined and cannot be skipped. | Compliance workflows, sensitive operations, data collection — where auditability matters more than flexibility. |
The trade-off is simple: determinism gives you control and consistency at the cost of flexibility. Dynamic reasoning gives you a natural experience with less predictability. Most businesses need a mix — this workshop shows you how to decide.
3. Bringing a human into the loop — at the right moment, through the right channel
An AI Agent that cannot hand off to a human is a dead end. You will configure escalation workflows that route customers to a real person based on different business conditions:
| Business Condition | What Happens | Channel |
|---|---|---|
| Customer asks to speak to a human | Agent stops and routes immediately — no friction, no persuasion to stay. | Phone or email, based on topic (e.g., security → phone, loans → email with callback). |
| Agent cannot answer the question | Agent acknowledges the gap honestly and connects to a specialist. | Asks certain questions before escalating and connects to a channel based on the business hours. |
| Topic is too sensitive for AI | Agent provides initial guidance, then hands off — regardless of whether it could answer. | Configurable per topic: phone for urgent (fraud), email for complex (large transfers). |
The principle: the customer never hits a dead end. The agent either answers, or connects them to someone who can.
4. Extending the agent beyond static articles with live tools
Knowledge articles cannot contain information that changes by the minute. In the workshop, the agent combines authored articles with live API calls — fetching real-time exchange rates, calculating tier-specific transfer costs, and looking up branch locations — to deliver answers that neither articles nor APIs could produce alone.
Session Details
| Duration | 90 minutes |
| Format | Hands-on, facilitator-led workshop with live audience interaction |
| Who Should Attend | Knowledge authors and content managers (primary). IT and CX professionals will also benefit. |
| Prerequisites | None. No technical or coding background required. |
| Materials | Everything is pre-loaded. Attendees do not need to bring or prepare anything. |
The One Thing Attendees Will Walk Away With
The AI Agent’s intelligence does not come from the model alone. It comes from how knowledge is written, structured, and tagged — and from knowing when to let the agent reason, when to lock it to verified answers, and when to step aside and bring in a human. Authors are not bystanders in the AI Agent era. They are the architects.
