Skip to main content
Every turn of the loop starts here: a gap in what your Customer Agent knows or can do. There are two ways to identify gaps:
  • Reactive: Prompt the Internal Agent for a report on how your Customer Agent is performing and where it falls short.
  • Proactive: Set up automations that watch your tools and your support queue and surface gaps on their own.
Either way, the Internal Agent reports the impact and proposes the update as a todo in a task. Each todo starts the next build, and the loop keeps turning.

​
How it works

1

Gaps surface

Reactively, when you prompt the Internal Agent for a report. Proactively, when automations watching GitHub, Linear, Slack, and your support queue catch a product change or a human-handled ticket - or when the Customer Agent flags missing context or a tool connection on a ticket (in beta - reach out and we’ll turn it on).
2

The Internal Agent proposes updates

It creates todos for the high-leverage improvements it identifies in automations and one-off tasks: knowledge gaps, tool ideas, and flow improvements. Each todo comes with the relevant context and attributes, assigned to the right user or team.
3

You start the todo

One click opens the task, and the next build begins. See Build.
Todos appear on the Tasks page and in your Inbox.

​
Reactive: prompt for a report

Prompt the Internal Agent in a task whenever you want to know how your Customer Agent is doing. It covers both the numbers and the why - including the impact of every release on your resolution rate.
  • Quantitative reports. “Show me a graph of resolved tickets over the past month” gets you an interactive chart, sortable by category. It can analyze any slice of your ticket data.
  • Qualitative reports. Patterns, trends, and representative quotes across your tickets. Common ones: performance reports, bug-pattern reports, and improvement reports.
A Customer Agent Performance report for the last 30 days, generated by the Internal Agent
When a report reveals a gap - a ticket type the agent keeps transferring, a policy it gets wrong - the Internal Agent proposes the fix as a todo in the same task.
Every ticket falls into one branch:
  • Engaged: Your Customer Agent replied to the ticket or took action on it.
    • Resolved: The agent closed the ticket, and it stayed closed for at least 3 days. Split into general, spam, and merged tickets.
    • Engaged, not resolved: The agent engaged, then handed the ticket to a human for final resolution (previously called “assisted”).
  • Not engaged: Your Customer Agent never worked the ticket.
    • Not assigned: The ticket was never assigned to the agent, for example because your routing workflow excluded it.
    • Ignored: The ticket was assigned to the agent, but it stayed out, for example because the ticket type is out of scope in its instructions.
Your Customer Agent also works as a team member in your ticketing system, so your existing reports track it automatically: SLAs, CSAT surveys, agent-level dashboards. We can also share report templates with the key metrics we recommend for every supported system.

​
Proactive: automations

An automation is an Internal Agent task that runs on a schedule or a trigger (see Automations). Set it up once, and it keeps watching for gaps - no prompt needed:
  • Your support queue: learn from tickets a human handled, and catch the questions your Customer Agent should have answered.
  • Your tools: a PR merging in GitHub, an issue moving to Done in Linear, a page changing in Notion - anything that means your Customer Agent needs an update.
  • Recurring reports: the same quantitative and qualitative reports, daily or weekly, with the gaps they reveal turned into todos.
When an automation finds a high-leverage improvement, the Internal Agent creates a todo with the relevant context and attributes, assigned to the right user or team. The Customer Agent flags gaps too: when it’s missing context or a tool connection on a ticket, it kicks off the Internal Agent to investigate. This is in beta and requires setup on our side, so reach out and we’ll turn it on for your workspace.
Each todo starts the next build. See Build.