> ## Documentation Index
> Fetch the complete documentation index at: https://docs.scaling.cloud/llms.txt
> Use this file to discover all available pages before exploring further.

# Work with AI findings

> Read, confirm, and dismiss findings on an incident, get a chat summary with @scaling, and approve a proposed action.

This guide covers the day-to-day of an [AI investigation](/concepts/ai-investigations) on a live incident: reading findings, weighing in, getting a summary in chat, and approving an action.

## Prerequisites

* A plan that includes AI Investigations (**Pro** or **Enterprise**).
* A `critical` incident (auto-run only fires at `critical` severity — you can also start a run by hand on any entitled incident).

## Read and weigh in on findings

<Steps>
  <Step title="Open the Findings panel">
    On the incident page, the **Findings** panel lists each finding with its `kind`, a one-line summary, a confidence percentage, and links to its evidence.
  </Step>

  <Step title="Open the evidence">
    Follow a finding's evidence link to the source — a past incident, a deploy, or a Sentry issue — to judge whether the diagnosis holds.
  </Step>

  <Step title="Confirm or dismiss">
    Click **Confirm** on a finding you agree with, or **Dismiss** on one that doesn't apply. This records the team's judgment; it does not change the incident's status.
  </Step>

  <Step title="Run again if the picture changes">
    As the incident develops — a new deploy, a fresh Sentry error — click **Re-run** to start another pass over the current signals.
  </Step>
</Steps>

<Note>
  Seeing no findings is meaningful. scaling.cloud only posts a finding when its
  confidence clears an internal floor, so an empty panel means nothing matched
  with enough confidence — not that the investigation failed to run.
</Note>

## Get a summary in chat

In the incident's Slack or Microsoft Teams channel, mention **@scaling** to get a live summary synthesized from the current findings, without leaving chat. The summary reflects the latest run, so mention it again after a re-run to refresh it.

## Approve a proposed action

Some findings propose a next step — file a ticket, post to a channel, attach a runbook. These appear in the **Proposed actions** panel.

<Steps>
  <Step title="Review the proposed action">
    Each proposed action names what it will do (e.g. *"Create a Linear issue"*). Open it to see the details.
  </Step>

  <Step title="Approve or dismiss">
    Click **Approve** to run it, or **Dismiss** to clear it. Nothing runs until you approve — the AI never acts on its own.
  </Step>

  <Step title="Confirm the outcome">
    On approval, scaling.cloud executes the action and records the result on the incident — for a create-ticket action, the linked ticket appears on the incident.
  </Step>
</Steps>

<Warning>
  A **create ticket** action needs your issue tracker connected. Connect
  [Linear](/integrations/linear) under **Settings → Integrations** first — see
  [Proposed actions](/concepts/proposed-actions#external-ticketing).
</Warning>

## After resolution

When you resolve the incident, its [post-mortem](/concepts/post-mortems) draft is pre-filled from the findings — root cause and contributing factors are seeded, so your retro starts from a draft instead of a blank page.