Deterministic End‑to‑End Testing for the AI Era.

!What happens

A coded check asks the page's code whether an element is there.Your user only sees the screen. These bugs pass a check of the code:

• Pay now

  We use cookies.OK

  Covered

  • The code: the button is there.
  • Your user: cannot see it, cannot press it.
• Order confirmed

  Invisible

  • The code: the text is there.
  • Your user: sees nothing.
• Confirm your order

  Cut off

  • The code: the full text "Confirm your order".
  • Your user: half a sentence.

• Shipping

  Payment

  Overlapping

  • The code: both cards are there.
  • Your user: a mess.
• HomeProductsContact

  Order confirmed

  Pay now

  Unstyled

  • The code: every element still there (the style sheet failed to load).
  • Your user: a broken site.
• Delete

  Wrong colour, broken picture

  • The code: the button and the image are there.
  • Your user: a different product.

A screenshot comparison can catch these too, at the price of a reference picture per browser and system, and false alarms when rendering differs between machines.

!What happens

A coded test points at the page's code.Rename a class, move a button into a new container, and the test fails although the screen looks the same.Playwright's own documentation warns that such locators "can break when the DOM structure changes".

The coded test clicks #pay-btn

!What happens

Coded tools find what to act on by its role, its label, its text or the app's code.Many apps draw their own pixels: a chart's bars, a map's pins, a game, a whole remote desktop.What is drawn has no element, role, label or text, so no locator can find it: the test clicks at coordinates, and a check needs the app's own code or a screenshot.

Most standard Windows apps (WinForms, WPF, WinUI, Qt, Electron) do describe themselves to Windows.GUI Test Studio uses that first, and finds the rest by sight.

!What happens

Every test is code, and every change to the app is a change to that code: the more tests a team has, the more code it has to keep working.The people who know what should be tested, testers and product owners, cannot write or fix it.

!What happens

A raw screenshot comparison can fail when fonts or rendering differ between machines: Playwright's own documentation names the operating system, its settings, the hardware and headless mode.

!What happens

A game draws its menus, inventory and HUD itself on the graphics card.Unity (6.3 and later), Unreal and Godot (4.5, experimental) can describe menus to screen readers only when the studio builds that in, screen by screen; the game world itself is never described.Coded tools also cannot play with a gamepad.

!What happens

Test screens show customer names, prices and internal tools.Visual testing services upload every screenshot to their cloud: Percy and Argos always, Applitools by default (its on-premises option is sold at a higher price).

!What happens

Every screenshot counts against a monthly allowance. Percy and Argos are limited to 5,000 screenshots a month on their entry plans; Percy's Desktop plan is $199 a month (billed yearly) for 10,000, then $0.036 per screenshot; Argos Pro is $100 a month for 35,000, then $0.004 per screenshot; Applitools starts at $667 a month (billed yearly).More browsers, devices and runs mean a bigger bill.

Prices as published by each vendor on October 28, 2026.

!What happens

Selenium and Playwright test web pages, WinAppDriver Windows apps, Appium phones: three tools, three skill sets, three kinds of test to maintain.

!What happens

A coded test fails with a stack trace, a page of code only a developer can read.