Skip to content

Build the Agent

If you finished Start Here, you already proved the first workflow: build, test, launch, and improve one version of Consultation Desk. This section picks up from that point and shows how an operator teaches the agent more clearly and shapes it into something more reliable across different touchpoints.

Consultation Desk in Agent Editor

What this section is for

This is not a tour of every setting. It is the working area for turning expertise into a reliable expert touchpoint:

  • define the job clearly in Editor
  • write better Instructions
  • test the behavior in Live Test
  • save progress with Apply
  • move the right version forward when it is ready

The running example

We keep using the same fictional agent from Getting Started: Consultation Desk.

We keep it because the core operator moves are easy to see. In this specific example, its job is to act as the front door for an expert service business:

  • welcome people who are not sure what they need
  • ask a few clarifying questions
  • recommend the right next step
  • hand off safely when the case is urgent, unclear, or outside scope

The allowed outcomes stay the same throughout this section:

  • Initial Consultation
  • Specialist Consultation
  • Human Callback

Do not read those exact outcomes as the product boundary. In another workflow, the approved outcomes might be feedback, a case summary, a safe escalation, or a structured next action.

Read this section in order

Page What you will learn
Agent Editor The day-to-day operator loop inside the editor
Instructions How to turn a vague brief into clear operating rules
LLM Model Selection When to keep the default model and when to switch
Version Management How Apply, notes, publishing, and rollback work together
Attachments When small always-on files help more than a large knowledge source

What stays out of scope here

This pass stays focused on how the agent thinks and behaves. Tool details such as Knowledge Base, Web Search, Call Agent, and Request Form still live in Tool Configuration.

Success for this section

By the end, you should be able to do three things confidently:

  • explain what your agent should and should not do
  • make one targeted change, test it, and save it cleanly
  • decide whether a problem should be fixed with Instructions, the model, an attachment, or a larger tool setup

Next Step

Start with Agent Editor.