Skip to content
AutoTestX Docs

AI in AutoTestX

AutoTestX uses AI in four places: to write test steps, to heal a step whose element moved, to diagnose a failure no rule explains, and to let your AI assistant run suites and explain results. In every one of them, a person decides: the AI drafts, suggests and explains, and nothing it produces changes a test until someone approves it.

What the AI does Where Turned on by
Write with AI Turns plain-language instructions into test steps The test editor Anyone who may edit the test, when the organization has an AI key
AI healing Picks which of the candidate elements is the one a broken step meant Runs, then the Self-healing page A project Admin or Test Manager (off by default)
AI diagnosis Reads an Unclassified failure’s evidence and suggests what went wrong The run report and suite runs A project Admin or Test Manager for automatic diagnosis; Diagnose with AI on one failure
AI assistants (MCP) Your assistant runs suites, reads the evidence of each failure and explains it Claude Code, Cursor, VS Code, Claude Desktop You, by connecting the assistant with an API token

Write tests with AI

Section titled “Write tests with AI”

Describe the test in plain language, such as “Sign in as the buyer, open a new purchase order, add one line and check the total”, and Write with AI drafts the steps in the open test. You review them and save. AI-written steps are saved as a Draft, so they do not run in a suite’s Ready-only gate until someone marks the test Ready.

  • Write with AI: how it works, what it generates and its limits
  • Prompt guide: what to tell the AI, with templates, and the master prompt for an AI chat that writes a complete test file
  • AI prompt examples: seven worked prompts and the tests they produced

Heal broken tests

Section titled “Heal broken tests”

When a button gets a new id or a label is reworded, self-healing finds the element the step most likely meant. Tiers 0 and 1 do it without AI. When Tier 1 has the right element in its list but no proof, the optional AI tier 2 is asked to pick.

  • The AI sees the step and up to six candidates (roles, names, labels, ids). Never the page itself, a screenshot, a typed value or a secret.
  • It answers with a candidate number, or none, and one sentence of reason. It never writes a locator.
  • An AI pick is never applied during a run, and rules never apply one. It waits on the Self-healing page for a person to approve, re-pick or reject.
  • Every decision is kept in the audit trail.

Diagnose failures

Section titled “Diagnose failures”

Every failure gets a failure class from rules: locator, assertion, server error and so on. When no rule matches, the failure is Unclassified, and the AI can read its evidence and suggest what went wrong.

  • The suggestion is labelled AI suggestion, with a confidence and the nearest category, the evidence it quotes, and up to three things to check.
  • It never changes the class: the failure still counts as Unclassified in Likely causes, in the release report and in anything a pipeline checks.
  • Automatic diagnosis is off until a project’s Admin or Test Manager turns it on, with a daily limit. Diagnose with AI on a single failure works either way.
  • Failures that look the same share one answer for 7 days, so twelve identical failures cost one.

On a suite run, Failure investigation groups the failures into Likely causes, so five tests failing on one endpoint read as one problem.

Work from your AI assistant

Section titled “Work from your AI assistant”

Connect your assistant to the AutoTestX MCP server once, then ask in your own words:

“Run the Checkout tests on Staging and tell me if anything broke.”

The assistant finds the suite, starts it, waits, and explains each failure from its evidence. In an editor it can also fix your code and run the suite again to confirm. It can also list the components that need maintenance. It signs in with an API token, so it can do only what that token’s role allows.

The AI key and what it costs

Section titled “The AI key and what it costs”

Write with AI, AI healing and AI diagnosis run on one Anthropic key per organization, chosen by an organization admin in Organization → Settings → AI key:

Choice AI runs on Counts against the plan’s AI allowance
Our own Anthropic key Your organization’s Anthropic account No
The AutoTestX key (the default, except on Enterprise) AutoTestX’s Anthropic account Yes
Turn AI off Nothing –

On the Enterprise plan, AI runs only on your own key. Each generation, healing pick and new diagnosis uses one generation from the allowance, and Organization → Settings lists the usage. See Your organization’s AI key.

An AI assistant connected over MCP uses its own AI, not the organization’s key.

Practise

Section titled “Practise”