Building a KATA Framework with Playwright

Most test frameworks fail because of architecture, not tools.

After building automation at scale for 5+ years, the pattern that consistently works is the KATA (Knowledge-Augmented Test Automation) framework: a three-layer architecture designed for both human and AI collaboration.

The 3-Layer Pattern

Layer	Name	Purpose	Owner
1	Component	What you test — page objects, API clients, UI components	Dev
2	Action	How you interact — reusable interaction patterns	Framework
3	Test	What you assert — scenarios and assertions	AI + Human

Layer 1: Components

Components encapsulate what exists in the system under test. They're the thinnest layer — just enough structure to identify elements and endpoints:

// Component layer — pure page objects
export class LoginPage {
  readonly emailInput = page.getByLabel('Email')
  readonly passwordInput = page.getByLabel('Password')
  readonly submitButton = page.getByRole('button', { name: 'Sign in' })

  async goto() { await page.goto('/login') }
}

Layer 2: Actions

Actions compose components into user-meaningful operations. This is where business logic lives:

// Action layer — reusable business flows
export class AuthActions {
  constructor(private login: LoginPage) {}

  async loginAs(email: string, password: string) {
    await this.login.goto()
    await this.login.emailInput.fill(email)
    await this.login.passwordInput.fill(password)
    await this.login.submitButton.click()
    await expect(page.getByText('Dashboard')).toBeVisible()
  }
}

Layer 3: Tests

Tests are the thinnest layer — pure intent, minimal implementation detail. This is where AI-generated assertions fit naturally:

// Test layer — intent-driven
test('user can sign in with valid credentials', async () => {
  const auth = new AuthActions(new LoginPage(page))
  await auth.loginAs('user@example.com', 'valid-password')
  await expect(page).toHaveURL(/dashboard/)
})

Why KATA Works for AI Collaboration

The architecture is designed so AI can contribute at any layer without breaking the others:

• AI generates Layer 3 tests from natural language: "test that a user can sign in" → the test above
• AI suggests Layer 2 actions from analyzing user flows
• Human owns Layer 1 components — element selectors require visual verification

The abstraction layers mean the AI can generate meaningful test scenarios without needing to know the exact DOM structure. And the human can update a component selector without breaking any tests.

Running It

npx playwright test

That's it. The framework handles parallel execution, reporting, and retries. KATA handles the maintainability.

Learn More

The full KATA implementation is in bunkai-qa-engineering — my open-source agentic QA platform. More details in an upcoming deep-dive.