How to Build Your Own AI Code Verification System >>> Join a hands-on webinar on Dec. 15th!

Home / Media & Entertainment Leader

Media & Entertainment Leader

On this page

An enterprise QA team moved off Katalon without losing coverage, then built an AI-assisted workflow that validates the Visual Quality of every new component against its design.

Highlights

Icon showing gear with lines connected to dots

Visual Quality coverage carried from Katalon to Playwright with the Applitools MCP Server, with no checks rebuilt by hand

Icon showing four interlocking puzzle pieces forming a square

4 design gaps caught in the first week, all while components were still in Storybook, plus 6 component variants validated against Figma from a single design URL

About the Company

The company is a large media and entertainment provider serving millions of customers across web, mobile, and connected TV. Its five-person QA team is responsible for the quality of customer-facing experiences where visual consistency directly affects the brand.

The Challenge: A Framework Migration and a Design-Heavy Release at the Same Time

The QA team had two projects on its plate that are usually planned separately.

First, they were moving their test suite from Katalon to Playwright. Standing up the new Playwright project was quick. The visual checks were the hard part. In most framework migrations, those get rewritten by hand, baselines get re-established, and coverage gaps open up along the way. With a release coming, the team couldn’t afford that.

Second, a dark mode release was coming. Components including global navigation, tooltips, and price blocks were deployed to Storybook, and each one had a Figma design defining how it should look. The team needed to confirm every implementation matched its design. Checking them frame by frame was not going to hold up across a full release.

On top of that, the team was part of a leadership-driven tool consolidation review, which meant every tool in the stack had to prove its value. 

“Setting up Playwright was the easy part. What we couldn’t afford was rebuilding every visual check by hand with dark mode on the way. The Applitools MCP server carried them over, so we started in Playwright with the same coverage we had in Katalon.” – QA Lead, large media and entertainment company

The Solution: Visual Quality Validation from Design to Code with Playwright, Claude, and Applitools

Migrating without losing coverage

The QA lead used Applitools MCP, which can configure Eyes and add visual checkpoints from the chat, to handle the Applitools side of the conversion. Instead of rebuilding Visual Quality checks in Playwright and re-establishing baselines, the team carried its existing coverage forward and moved straight on to new work.

Making Figma the source of truth

The design validation workflow has three parts:

  • Claude writes the tests and code changes. Code changes on the team now go almost entirely through Claude.
  • Applitools MCP server gives Claude direct access to Eyes. Claude uses it to run Visual Quality checks, get tests passing, and rule out failures caused by flaky tests rather than real defects.
  • Figma Design Baselines links each test to its Figma frame or component, and that design becomes the baseline.

The team’s previous approach relied on exporting designs through a plugin, one frame at a time, by hand. Rolling that out to the rest of the QA team would have meant sharing Figma credentials, since the company licenses only a few Figma seats. It would also have meant introducing a new practice right before launch. With Figma Design Baselines, no plugin is required. The team pastes the Figma URL into the Playwright configuration next to the test it belongs to. Eyes pulls the design, uses it as the baseline, and sizes the test viewport to match. In one run, a single URL picked up all six variants of a component. Each result in the Eyes dashboard shows the design’s name, type, and file revision, with a link back to the exact frame it was compared against.

The workflow continues after a component ships. Once an implementation is approved, it becomes the baseline for regression testing. If a designer later updates the Figma file, Eyes detects the change and compares the implementation against the new design, so no one has to track revisions or re-baseline manually. The workflow reads each story and its Figma links from Jira, and test results report back to Zephyr. Findings are tagged with an “AI generated” label, so the team can see which issues the workflow surfaced.

“Before, every frame was a manual export, and getting the rest of the team on it would have meant sharing Figma logins. Now it’s a URL in the config. One link picked up all six variants of a component. The only manual step left is supplying the design links, and our UI/UX team is fixing that by adding component-level links to every story.”

Why not LLM vision alone?

The team did not arrive at this combination by assumption. As part of the consolidation review, they compared Applitools against LLM vision on their own pages to see whether a model could handle Visual Quality validation by itself. It could not. LLM vision returned different verdicts on identical images, had no reliable answer for dynamic content, and covered none of the parts of Visual Quality validation that happen over time. It also lacked the platform capabilities the team depends on:

  • Saved baselines and a shared review dashboard
  • Full-page capture and stitching
  • Root cause analysis down to the DOM and CSS
  • Region annotations and match levels for dynamic content
  • Baselines that follow branches through source control, with a gate in CI

Scale was the other gap. The dark mode components had to look right on every browser, device, and viewport the company’s customers use. With the Applitools Ultrafast Grid, the team renders each component across all of them from a single baseline, without maintaining its own grid infrastructure. Doing the same with LLM vision would mean reviewing every rendering separately, multiplying both the token cost and the chances of an inconsistent verdict.

Applitools Visual AI is deterministic and built for exactly those jobs, which is why the workflow uses both rather than choosing one.

Keeping the model focused

Running an LLM alongside Visual Quality validation raises a fair question about cost. The workflow is built to keep the model focused. Visual AI identifies what actually changed and hands the model targeted evidence, such as clustered diff regions, DOM changes, and cropped images, instead of full screenshots. The team gets the reasoning benefits of AI without the cost and noise of making a model sift through entire pages.

The Results: Design Gaps Caught While Components Are Still Being Built

In a single week, the workflow flagged four issues across three stories: two in global navigation, one in a tooltip, and one in the tab tray. Each was caught while the component was still in Storybook, ahead of the dark mode launch, and routed into the Jira and Zephyr processes the team already uses. The tab tray finding turned out to be a late requirement change rather than a defect, but it still exposed a real gap between design and build that might otherwise have surfaced much later.

  • 4 design gaps caught across 3 stories in one week
  • Components validated against Figma include global navigation, tooltips, price blocks, search and cart components, and a component with six variants

“We caught four design gaps in the first week, all while the components were still in Storybook. That’s when a fix is quick. Finding them before launch instead of after is the whole point.”

A Path Forward: Toward a Fully Connected Workflow

The next step is removing the last manual piece. The QA lead plans to have the UI/UX team include every relevant component-level Figma link in each story, not just a page-level one, so the workflow can pull design links on its own. The QA lead runs the workflow on all of his stories, and other QAs on the five-person team have already tested components with it. Two have Claude licenses today, the other three work through them, and the team plans to add more licenses.

What began as a framework migration has become a different way of working. Design, code, and Visual Quality validation are connected from the start, and the team learns whether a component matches its design while it’s still being built, when a fix is simple and long before a customer sees it.

©2026 Applitools