Understand AI Agents
An AI Agent is a reusable conversation configuration: its purpose, instructions, knowledge, active hours, allowed actions, tags, and routing. One account can use different agents for different offers, audiences, or workflows.
Give each agent one clear job
A reliable AI Agent combines approved business knowledge with clear behavior, routing, availability, and handoff rules. Keep each agent focused enough that your team can tell whether its answer and next action are correct.
What belongs to an AI Agent?
An agent is more than a prompt. It is the controlled operating configuration that shapes how defined conversations are handled.
Instructions
Role, tone, goals, questions, response style, boundaries, and escalation rules.
Knowledge
Approved FAQs, pages, documents, and maintained business sources.
Active hours
The schedule that controls when the agent may process new inbound messages.
Abilities
Approved actions such as collecting details, creating a task, or offering an appointment.
Tags
Labels the agent may add or remove when the approved conditions are met.
Entry points
Rules that decide which new conversations reach this agent and which do not.
From a new message to the right next step
The entry point provides context, the agent applies its approved configuration, and the conversation reaches the expected inbox, action, or human handoff.
Use one coherent job per agent
A broad agent that mixes unrelated offers, policies, audiences, and next actions is harder to test and more likely to use the wrong context.
When the operating context stays consistent
- The same knowledge and policies apply
- The qualification questions stay the same
- The same team owns the handoff
- The same next action remains appropriate
When the job changes materially
- The offer or audience needs different facts
- The tone or language changes
- The operating hours or handoff team differs
- The desired outcome uses another workflow
Build the complete operating configuration
Work through these guides in order for a new deployment, or open the guide that matches the part of an existing agent you need to review.
Connect each traffic source or trigger to the AI Agent prepared for that context.
Define the role, voice, questions, boundaries, next actions, and human handoff rules.
Give the agent clear, current, approved information without mixing unrelated sources.
Control when the agent may process new inbound messages and plan coverage outside those hours.
Understand how the first inbound message is evaluated and how to correct a false positive.
Turn an agent off without deleting it
Turning an agent off stops it from replying through its assigned routes. It does not delete the configuration or conversation history, and the connected channel may remain active.
Confirm queue ownership before disabling an agent so incoming conversations are still reviewed by a person.
Only one conversation needs a person? Pause the AI for that contact instead of turning off the agent for every route.
Make changes safely
- 1
Identify the exact scenario
Record the conversation, the observed response, and the behavior your team expected.
- 2
Change the smallest relevant source
Adjust the instruction, knowledge item, route, or schedule responsible for the result.
- 3
Retest the original and related cases
Confirm the correction without weakening a nearby answer, boundary, or exception.
- 4
Verify the complete journey
Check routing, human handoff, contact updates, and every enabled action affected by the change.
- 5
Approve and monitor
Record the accepted result and review the next live examples for unexpected behavior.
Before an agent receives customer traffic
- The agent has one defined job, audience, and appropriate next action.
- Instructions, knowledge, active hours, entry points, and abilities are approved.
- Uncertainty, sensitive requests, and requests for a person have explicit handoff rules.
- Normal, ambiguous, out-of-scope, and adversarial scenarios have been tested.
- The receiving team can find, own, and respond to a human handoff.
- A person is assigned to review early conversations and approve corrections.
Start with the conversation outcome
Send us the traffic source, audience, customer questions, desired next action, and handoff owner. We can help determine whether the use case belongs in one agent or several focused agents.