AI agent overview

What an Interlinked AI agent is

An AI agent is the customer-facing assistant that represents your business in the channels you assign to it. Its behavior comes from the business context, workflow, knowledge, catalog or service information, voice, abilities, and governance choices you review in the panel.

Open My Agents in the AI & Automation section to see the agents in your workspace. The agent list shows whether an agent is a draft, published, active, or turned off, and whether it still has launch requirements to complete.

The agent lifecycle

Interlinked keeps the main lifecycle actions separate so you can review changes before they affect customer conversations:

1. Create a draft — choose the business context, the agent’s job, an AI model, and the required operational information.
2. Configure the draft — refine the workflow, connections, catalog, voice, abilities, knowledge, and governance.
3. Publish the configuration — make the reviewed draft the agent’s current live configuration.
4. Activate the agent — validate remaining requirements and turn the agent on for an assigned customer channel.
5. Pause or resume — turn the agent off when you need a break; its configuration and published versions remain available.

Publishing and activation are different actions. Publishing does not automatically turn the agent on or off, and an agent can be published while it still needs a connection, model, or other requirement before it can launch.

What shapes the agent’s behavior

The most important customer-facing inputs are:

• Identity and business context — who the agent represents and the rules it should follow.
• Workflow — the job the agent performs, such as product conversations, service requests, appointments, qualification, or handoff.
• Model and voice — how the agent reasons about a request and how it communicates.
• Apps and channels — the connected services and customer channels available to the agent.
• Catalog and knowledge — the products, services, and information the agent can use when answering.
• Abilities and governance — actions the agent may perform, boundaries, and places where a person or approval is required.

For a guided first setup, start with Create an AI agent. For an existing agent, use Configure an AI agent.

Draft changes and the live agent

When you edit an existing agent, the panel identifies unpublished changes. The live agent keeps its current configuration while you review the draft. Open Review & Publish to inspect the pending changes, or use Discard Changes to reset the draft to the live baseline when that option is available.

After publishing, the agent still follows its current on/off state. Read Review agent readiness and Publish and activate an AI agent before launching it to customers.

Conversation control after launch

Once an agent is active, monitor customer conversations in CRM. You can take manual control of a conversation when a person needs to respond. A manual reply on Instagram, Facebook Messenger, or WhatsApp pauses AI handling for that conversation for 15 minutes after the latest human message; the timer starts again when another manual message is sent. See Switch between AI and manual control.