Agentic testing for the AI era

Your coding agent writes the feature.
Let it test it too.

Automatically plan and run frontend tests in a real browser, discover issues, and fix them—with complete execution logs for review.

Works with your favourite coding agent

Download for Windows Download for Linux

RealBrowser based testing

AutomatedAgent-driven testing

ShareableExecution reports

ReliableReproducible runs

Checkout flow

Execution log

Step-by-step record of this execution.

Navigated to checkout
Filled shipping and selected saved card
PassAssertion: Saved card details are autofilled
POST/api/charge200

Automated Testing

Agent driven test planning, authoring, and execution

The agent drafts the plan, writes the tests, and runs them in the browser.

Test planning

From your feature and repo context, the agent drafts what to test — core paths, edge cases, and failures.

Come up with the test scenarios
Discover corner cases
Simulate error scenarios

Test plan

Scenarios being tested in this session.

Checkout — happy path
Signed-in user with items in cart completes purchase with a saved card; order confirmation renders.
1. Sign in as test user with cart pre-filled 2. Open checkout and select saved Visa ending 4242 3. Confirm order and wait for success route 4. Assert confirmation id and email match session
Cart — empty checkout blocked
Checkout CTA stays disabled when the cart is empty; no network call to payment intent.
Payment — decline and retry
Simulate card decline, surface inline error, allow user to change method and succeed.

Test authoring

The agent writes tests in your project, covering all the scenarios from your test plan.

AI written tests
Implement scenarios through mocking
Re-usable tests for regression testing

checkout.js

await page.goto(url + "/checkout");

await record({

id: "checkout-open",

message: "Checkout loaded",

attachments: [{ kind: "screenshot", buffer: await page.screenshot() }],

});

await page.locator('[data-testid="checkout-pay"]').click();

await expect(page.locator('[data-testid="order-confirmed"]')).toBeVisible();

Test execution

Agent runs the tests, and generates execution logs for review.

Automatic test execution
Execution logs for review
Feature wise testing session

Execution log

Step-by-step record of this run.

Opened checkout with items in cart
http://localhost:3000/checkout
Checkout3 items
- Canvas weekender
  $48
- Sticker pack + bottle
  $44
Subtotal$92.00
Continue
Filled saved Visa ending 4242
Submitted order — confirmation route loaded
Assertion passed: order id and email visible

Debugging

Automated debugging and fixing of test failures

Let the agent debug and fix issues; jump in to assist with tricky scenarios.

Agent sees the issue and fixes it

When a test fails, the agent can spot what broke and fix it.

acme-frontend

> /lh-testing Test the checkout success flow

I'll invoke the subagent to run that test in the browser session and inspect the failure output.

longhorizon.runTest({ testId: "checkout-submit", profile: "normal" })

FAIL: inventory returned checkout blocked (expected allow)

The run failed because the inventory blocks checkout. I'll patch the saved test to stub inventory with 3 units, then ask the subagent to run again.

Edit(.longhorizon/tests/checkout.js)

Re-invoking the Long Horizon subagent with the same test after the mock fix.

longhorizon.runTest({ testId: "checkout-submit", profile: "normal" })

PASS · Checkout — submit when stock available

>

? for shortcuts

Developer reviews execution logs

Agents make mistakes. Developers can review the execution logs to point out any oversights.

Execution log

Step-by-step record of this run

Navigated to /checkout
Filled shipping — US, 94107
Clicked Pay — confirmation route
Assertion: order id visible in DOM

Slow mode and step mode

Run the test in slow mode or step mode to manually try any scenarios.

http://localhost:3000/checkout

UI Feedback

Feedback interface built for optimal agent collaboration

Leave feedback directly on your UI so agents can implement changes faster.

Leave comments on the UI elements for the agent to implement.

Provide comprehensive feedback on a screen without multiple back-and-forth.

Agent automatically gets the screenshot, comments and the element HTML.

/checkout

Checkout

Discount coupon

SUMMER-15 applied

Add border and shadow

Payment method

Credit card ending in 4408

PayPal

Increase row spacing

Use primary button style

UI Feedback

Terminalclaude

On the checkout page, make the following changes:

1. Add border and shadow to the discount coupon section.

2. Increase spacing between payment method rows.

3. Update Place Order to a stronger primary style.

Working...

http://localhost:3000/checkout

Checkout

Add border and shadow

Discount coupon

SUMMER-15 applied

Increase row spacing

Payment method

Credit card ending in 4408

PayPal

Summary

Subtotal$168.00

Shipping$6.00

Total$174.00

UI Feedback

Terminalclaude

Checkout page implementation is done. All the tests passed and lints are clean.

Marking the last TODO item as done.

On the checkout page, make the following changes: 1. In the discount coupon section, add a border and soft shadow to the card. 2. Increase spacing between payment method rows for easier scanning. 3. Update the Place Order button to a stronger primary style.

I'll parse each item and figure out where these elements live in the UI.
Let me check the current implementation of coupon card, payment list, and checkout CTA.

Working

PR Review

Review PRs with confidence, not guesswork

AI has accelarated development velocity, but manual review process is breaking. Approve PRs with confidence after you see evidence that the feature works.

//Pull requests#1842

Openalex wants to merge 4 commits into main

feat(checkout): telemetry, validation, and regression coverage

Conversation

alexcommented 2 hours ago

Summary

Adds declined-card telemetry, tightens checkout validation, and updates mocks for the payment sandbox.

All success and error paths are tested. Regression testing is also done. Please see the Long Horizon testing report.

Long Horizon testing report

https://longhorizon.dev/reports/k8n3xq

What was tested

Reviewers can see which test cases ran and point out any gaps.

Execution logs

Reviewers can check execution logs to see how the feature works.

Evidence-backed approval

Approve PRs with confidence backed by concrete run evidence.

Execution Reports

One link, the full story

Each testing session generates a shareable report. Link the report with the pull request to elicit confidence from your reviewers.

Checkout flow

Test plan

Scenarios being tested in this session.

Checkout — happy path
Signed-in user completes purchase; confirmation renders with order id.
Cart — empty checkout
Checkout stays disabled with an empty cart.
Promo — stackable coupons
Discounts combine without double-charging shipping.
Guest — checkout without account
Email-only flow still enforces fraud checks before payment.

Compatibility

Works with all coding agents

Stay inside your coding agent and use your favorite models. Long Horizon integrates seamlessly with the tools you already use.

Claude Code

Anthropic's coding agent

Cursor

AI-first code editor

Codex

OpenAI's coding assistant

and many more — any MCP-compatible coding agent works with Long Horizon

Welcome to the world of agentic UI testing.

Get started with Long Horizon today.

Download for Windows Download for Linux

Your coding agent writes the feature.
Let it test it too.

Test plan

Execution log

Agent driven test planning, authoring, and execution

Test planning

Test plan

Test authoring

Test execution

Execution log

Automated debugging and fixing of test failures

Agent sees the issue and fixes it

Developer reviews execution logs

Execution log

Slow mode and step mode

Feedback interface built for optimal agent collaboration

Review PRs with confidence, not guesswork

feat(checkout): telemetry, validation, and regression coverage

What was tested

Execution logs

Evidence-backed approval

One link, the full story

Test plan

Tests which were run

Human readable summary

Execution logs

Test plan

Execution log

Works with all coding agents

Claude Code

Cursor

Codex

Welcome to the world of agentic UI testing.

Your coding agent writes the feature.Let it test it too.

Agent driven test planning, authoring, and execution

Test planning

Test authoring

Test execution

Automated debugging and fixing of test failures

Agent sees the issue and fixes it

Developer reviews execution logs

Slow mode and step mode

Feedback interface built for optimal agent collaboration

Review PRs with confidence, not guesswork

feat(checkout): telemetry, validation, and regression coverage

What was tested

Execution logs

Evidence-backed approval

One link, the full story

Test plan

Tests which were run

Human readable summary

Execution logs

Works with all coding agents

Claude Code

Cursor

Codex

Welcome to the world of agentic UI testing.

Your coding agent writes the feature.
Let it test it too.