Create your first AI Agent — and fix it when it doesn't run

How agents work, how to build one in the chat builder, approval vs autosend, and a troubleshooting checklist.

Agents are K3X's AI workers. Each agent has a goal, a set of tools it's allowed to use, and triggers that wake it up. You describe the outcome in plain language; the agent executes it and adapts along the way.

Build one

  • Go to /agents/new. The chat builder asks one clarifying question at a time.

  • Describe the goal the way you'd brief a teammate, for example: "When a new demo request comes in via webhook, create a lead with the form data, assign it round-robin to the sales team, and draft a welcome email for me to review."

  • Pick the trigger(s) that should wake the agent (see below) and confirm the tools it may use.

Trigger types

  • New lead created

  • Deal hasn't moved in a set number of days

  • Email arrives on a lead

  • On call end / missed call / SMS received

  • On a schedule (e.g. every morning)

  • Webhook — an external system POSTs to the agent's unique URL

Drafts vs autosend

By default, agents draft emails and SMS and wait for your approval before anything goes out — AI never sends without you unless you explicitly opt into autosend for a specific agent and channel. The mental model: AI Chat always drafts; agents are where direct send lives, and only when you turn it on.

Pausing and turning off

Open the agent and switch it off to stop new runs. If you're seeing activity after turning an agent off, check two things: runs that started before you switched it off may still complete, and some emails may actually be notifications about agent activity rather than agent-sent emails — see "Notifications: what K3X sends you and how to control it." If something still doesn't add up, report it from the widget; we take runaway-agent reports seriously.

Troubleshooting checklist

  • Agent never fires. Is the agent switched on? Is the trigger the one you meant (e.g. "new lead" vs "webhook")? For webhook triggers, confirm the sender POSTs JSON to the exact URL — a test with curl or Zapier's test button should show a run within seconds.

  • Agent fired but did the wrong thing. Open the agent's run history to read what it saw and did, then tighten the goal wording. Concrete rules ("only leads from the US", "never email existing customers") work better than vague ones.

  • Agent ran but nothing was sent. It's probably waiting for your approval — check the drafts queue on the lead or your notifications.

  • Runs failing. The run log shows the error. Signature mismatches (webhooks) and missing fields in the payload are the usual causes.

Agents pull from your AI credit pool per run — see "How credits work in K3X."

Did this answer your question?
😞
😐
😁