AI Assistance

DontBreak uses AI to take the busywork out of test maintenance: writing clear step descriptions for you, and explaining why a run failed so you don't have to scrub through video. This page covers what's available today, plus a preview of generation features currently in early access.

AI Descriptions

Every step you record gets a default machine-generated name. AI Descriptions rewrites your test's title, description, and per-step descriptions into clear, human-readable language based on what the steps actually do.

You'll find it in the visual editor under Test → AI Descriptions. The editor also offers it proactively: when you save a test whose steps still have default descriptions, DontBreak asks whether you'd like better ones generated first.

How it works:

  1. Open Test → AI Descriptions in the editor.
  2. Click generate — the AI reads your steps and proposes a test title, a test description, and a description for every step.
  3. Review the proposed text, then apply it to the test.

Good descriptions aren't just cosmetic: they make reports readable for teammates, and they give the AI failure analysis (below) more context to reason with.

AI failure analysis

When a run fails, DontBreak's AI analyzes it automatically, as soon as the run's screenshots are uploaded. It examines the failing step, its screenshots, the errors, the results across the browsers the run used, and the test's recent runs, and returns:

  • An issue type classifying the failure (e.g. an app bug vs. a test problem).
  • A confidence score for the diagnosis.
  • A plain-language explanation of what went wrong and what to do about it.

The analysis appears at the top of the report page. The recent runs let it tell a step that fails on and off (usually timing) from one that passed for weeks and then broke (usually a change in your app).

In alerts. Failure alerts by email and Slack include a one-line Likely cause when the analysis is confident. The alert waits for the analysis, which usually adds under a minute; if it takes longer than two minutes, the alert goes out without it.

Repeat failures. A run that fails the same way as an earlier one (same steps, same browsers, same errors) shows the earlier analysis instead of a new one, and the report says so. Click Re-analyze on the report for a fresh look.

Turning it off. Automatic analysis sends the failing steps' screenshots to Anthropic's Claude API. The team owner can switch it off under Settings → Team & Members → AI. With it off, the report shows an Analyze failure button instead, and nothing is analyzed until someone clicks it. See Understanding Test Results and Common Errors.

Early access: AI test generation

A set of larger generation features is rolling out gradually and is not yet generally available. You may see them in your account if you're in the early-access group:

  • AI Test Generator — describe a flow (or point at a URL) and get a complete draft test generated for you.
  • Test Case Generator — on a test suite, discover and generate test cases for the site the suite covers.
  • Suggest next step — a guided flow in the editor where the AI proposes the next step based on the page and what your test is trying to accomplish.

Availability

These features are experimental and their behavior and UI may change before general release. If you don't see them in your account yet, they haven't been enabled for your team.