Skip to content

Start Here

If you are a first-time buyer or operator, this is the recommended first path.

Codeer helps an expert team bring judgment into customer conversations and workflow moments. Instead of building a scripted chatbot or shipping a prompt on faith, you build the agent and the verified scenario set together: what it should answer, what it should not answer, and what it should hand off.

In this getting started flow, you will build a simple customer service agent around a narrow first scope. The example is intentionally plain so the method is easy to see: start from the questions you want to support, test the boundaries, and keep everything else on a safe fallback until a later version.

This guide takes you through the first complete loop: discuss the goal with Copilot, build the first agent, ask Copilot to generate the first important scenarios, immediately test 20 to 30 of them in Test Suite, then fix the cases that do not pass.

Once you reach that point, you no longer have an empty agent whose behavior is hard to see. Even if the score is not perfect yet, you can see how the agent behaves in the scenarios that matter. Finding the problems is the first real step toward controlling them.

Customer service agent running in the current Agent Editor and Live Test layout

What You'll Build

By the end of this guide, you will have:

  • A first version of a customer service Agent created with Copilot
  • A first scenario set covering core, boundary, and action scenarios
  • Reusable Test Suite cases with checkable Standard criteria
  • A published Web channel with a shareable slug-based URL
  • Stakeholder feedback collected from real conversations
  • A clear next set of scenarios to add in a later version

What this customer service agent should do

Your first customer service Agent should:

  • Answer a small set of customer questions that have been thoroughly tested and verified
  • Refuse, hand off, or request a form when the question is outside the first scope
  • Avoid inventing policies, prices, guarantees, or outcomes that your team has not approved
  • Use only the knowledge and tools needed for the first verified scenarios

Start with important scenarios

Before you expand the agent, define the first scenario set you truly care about:

  • Core scenarios the agent should answer
  • Boundary scenarios it should refuse, hand off, or route to a form
  • Action scenarios that need a tool

Read Verified Scenarios when you want the full method for defining and expanding trusted scope.

Prerequisites

Before you begin, make sure you have:

  • A Codeer.ai account
  • Access to a workspace where you can edit Agents and publish channels
  • A modern browser such as Chrome, Firefox, or Edge
1Set Up Workspace
↓
2Define Verified Scenarios
↓
3Build and Debug First Draft
↓
4Verify Scenarios in Test Suite
↓
5Publish Stable Scope
↓
6Review Production Feedback
↓
7Expand by Version

Step 1: Set Up a Workspace

Create a clean place for this pilot so your Agent, channel, and feedback stay together.

→ Set Up Your Workspace

Step 2: Define Important Scenarios

Ask Copilot to generate the first important scenarios from the intended first scope. Include the core questions the agent should answer, plus boundary and out-of-scope questions.

→ Verified Scenarios

Step 3: Build and Debug the First Draft

Use Copilot to draft the first customer service Agent. Keep the knowledge base as small as possible and include only what the first important scenarios need.

Use Live Test for quick draft debugging, but do not treat it as the release gate.

→ Build the First Version

→ Build and Debug First Draft

Step 4: Verify Scenarios in Test Suite

Turn the first scenario set into reusable cases and immediately run 20 to 30 of them. Review each AI response against its Standard with Copilot, find the cases that do not pass, then fix the draft until the important cases are stable.

→ Verified Scenarios

Step 5: Publish Stable Scope

Publish the Agent and Web channel only after the first verified scope is stable. Questions outside the first version should be refused, handed off to a human, or routed to a form.

→ Publish Stable Scope

Step 6: Review Feedback and Expand

Review what real users asked in Histories. Even when the first version refuses or hands off out-of-scope questions, those questions are valuable because they show which scope may be worth expanding next.

→ Review Feedback and Expand

Next Steps

After this first verified launch loop, continue with: