Playwright gives you fast, reliable execution. AI layers can suggest locators when the DOM shifts—but only if your suite is structured so recovery suggestions are safe to apply. This article walks through a pattern we use with enterprise teams running hundreds of UI tests in CI.
Marcus Chen
Test Automation Lead · 5 March 2024
Playwright gives you fast, reliable execution. AI layers can suggest locators when the DOM shifts—but only if your suite is structured so recovery suggestions are safe to apply. This article walks through a pattern we use with enterprise teams running hundreds of UI tests in CI.
01
Prefer getByRole, getByLabel, and getByTestId over CSS paths that break when design tokens change.
02
Keep navigation, forms, and assertions in separate modules so a healed locator touches one file.
03
Attach screenshots, trace IDs, and DOM snapshots when recovery fails so humans triage quickly.
Recovery algorithms work best on small, isolated surfaces. When every test reaches through a monolithic page object, a single healed locator can mask unrelated regressions.
“The maintenance tax drops when tests describe user intent—not DOM archaeology.”
// pages/checkout.page.ts
export class CheckoutPage {
constructor(private page: Page) {}
promoField() {
return this.page.getByRole('textbox', { name: /promo/i });
}
async applyPromo(code: string) {
await this.promoField().fill(code);
await this.page.getByRole('button', { name: 'Apply' }).click();
}
}Signal | Action | Owner |
|---|---|---|
Locator not found | Suggest role/label alternative | AI assist + dev review |
Assertion timeout | Capture trace, do not auto-heal | QA triage |
Flaky pass/fail | Quarantine + root cause | Platform team |
| Context | A data grid column header rename breaks forty tests that used visible text selectors. |
|---|---|
| Challenge | Blind auto-heal would map text to the wrong column when labels are similar. |
| Solution | Recovery runs only inside the grid component boundary, prefers aria-sort and column keys, and opens a PR comment when confidence is below threshold. |
| Outcome | Thirty-four tests self-heal; six require human update due to ambiguous labels. Mean time to green drops from six hours to forty minutes. |
Self-healing is a maintenance strategy, not a substitute for design-system stability. Invest in test IDs for dynamic components, keep recovery scoped, and review healed diffs the same way you review application code.