Agent Examples & Use Cases

What You'll Learn

Walk through three real-world agent examples and see how configuration changes based on the use case. Different agents share the same underlying principles, but need different prompts, tools, data, memory, and evaluation criteria.

Example 1: Sales Roleplay Agent

Practice cold calls against a realistic AI buyer that pushes back.

AI Instructions — Key Elements:

• Character setup — Give the AI a persona (name, title, company, existing tools) so it behaves like a real prospect
• Resistance patterns — AI is trained to be helpful by default; you must explicitly instruct it to give pushback and even come across as rude at times
• Breakthrough conditions — Sales doesn't happen gradually; define the specific criteria (pricing concern addressed, urgency created, problem acknowledged) that unlock a sale
• Core objections — List the objections the AI should raise so the caller practices handling them
• Early termination — If the user resorts to hard-sell tactics or misbehaves, the agent cuts the call

Rubric:

• Opening — Did the caller introduce themselves clearly and earn time?
• Rapport building — Was a connection established before pitching?
• Objection handling — How well were the buyer's concerns addressed?
• Closing — Was there a clear next step or ask?

Example 2: AI Cold Calling Agent

Automate outbound lead qualification calls at scale.

Script Structure:

• Phase 1 — Opening — The richer the opening details, the less likely the lead hangs up; deliver your value prop in the first beat so they give you five minutes
• Phase 2 — Qualification — Discovery and selling go hand-in-hand; the agent can't just ask questions — it must add value, prime the lead, and give cues ("That timeline is very favorable, our team could help")
• Phase 3 — Q&A — Let the lead ask questions before closing, otherwise the call feels rushed
• Phase 4 — Closing — Wrap up with a clear next step

Configuration Tips:

• FAQ guidance — Small FAQ sets can go in the system prompt to avoid the 300–500ms knowledge-base retrieval latency
• Variable extraction — Enable this to capture structured lead data (qualification answers, contact info) from the conversation
• Conversation strategy — Add filler words for natural speech; ~5 seconds of talkativeness is optimal; enable Conductor to let the agent cut abusive calls
• Iteration is essential — Cold call scripts almost always need tweaking after real calls; a script that works for humans may not work for AI due to delivery differences

Example 3: AI SDR with Browser Tool

Let an AI agent demo your product live using authenticated browser sessions.

Building the Demo Story:

• Define a multi-phase story the agent follows (e.g., homepage → create page → feature walkthrough)
• For each phase, specify which page opens, what to capture, and what to explain

Browser Tool Commands:

• goto — Navigate to a specific URL
• act — Click elements or perform actions on the page
• capture — Take a screenshot of the current view

Setup:

• Enable the Browser tool in the Tools section
• Click "View Session" to authenticate — stored cookies are reused for all future sessions

Key Differences from Other Agents:

• Conversations are longer (~10+ minutes) because users are warmer — they came to your website
• Getting the demo story right is the most critical part
• Can also be used for guided support — walk users through actions they struggle with
• Buyer signal tracking and rubrics work similarly to cold calling agents