When QA hits a wall, can AI build its way through?

QA is one of the quietly critical moments in any build. It’s where you find out whether the work holds up under real conditions. What happened on a recent build caught the room’s attention at this month’s Meet Share Question. Matt Hollander, a Product Owner at MSQ DX, took the prize for showing how AI could step in when QA hits a wall.

Meet Share Question is our monthly AI show and tell. Once every two months, four colleagues take the floor to share what they’ve been doing with AI, whether it’s solving a problem in their own work or something the wider business could pick up and run with.

The problem AI was asked to solve

Matt’s team looks after the website for a UK holiday-park operator. The build pulls data from a third-party API and surfaces relevant events, facilities and activities for people searching for holidays. The kind of integration that makes the experience feel personal.

When the build hit QA, our test team needed real data to put it through its paces. Specific dates, specific scenarios, the edge cases that only show up when you throw live information at a system. That data needed to come from external systems, and getting it at the right level of detail was going to take time the team didn’t have.

Rather than wait, Matt asked a different question. What if AI could help the team build the test data tool itself?

What he made

A QA Test Data Finder. A scrappy tool that pulled live data through the API and surfaced exactly the scenarios the QA team needed.

What stuck with the room was that it wasn’t one perfect prompt, six tools did the work between them. Claude Desktop generated the code, while project memory and the ADO MCP gave it the context, pulling in client history, user stories and earlier wiki notes. The Knack API spec defined the data contract, the QA team’s scenarios defined what good looked like, and the Filesystem MCP handed over the codebase.

Six inputs, three hours on a train to Leeds, one tool. Each of those tools is fairly ordinary in isolation. The work is in choosing them, sequencing them, and giving the model enough context to do something useful.

The hour he set on fire

What made the talk land wasn’t the polished outcome, it was the bits that went sideways along the way. Matt spent an hour fixing a script that was already running, because the UI he’d let the model generate on top was silently failing to render and he’d assumed wrong about which thing was broken. Easy to go overboard, as he put it. The script had crept into a UI.

He also kept finessing past the point of MVP, and Claude, for the record, started the session “with some sass.” Matt had screenshots, in case anyone still doubted that AI has a personality.

There’s something in hearing someone who’s just shipped a working tool tell you about the hour they wasted, because that’s the bit most of us recognise from our own work.

The unexpected upside

Once it was running, the tool did more than asked. Beyond solving the test data problem, it quietly exposed four real bugs in the build. It also surfaced performance issues in the third-party API. Busy-period requests were timing out, which nobody had caught because nobody had been able to throw real test data at it before.

Cue meetings and findings shared with both client and third party. By the time Matt presented, the third party had made API changes to support the use case. The team is now scoping follow-up work with a release in sight.

AI didn’t replace the QA process here. It unblocked it. It let an engineer with product knowledge see the issues earlier, in sharper detail, and get them in front of the right people sooner.

So why did this one win

Every Meet Share Question reminds us how broadly AI is showing up across the business. April was a great example. Emma and John walked us through a personalisation engine they’ve built for a luxury hospitality client, with seven specialist agents wrangling a 6,384-point signal matrix. John came back solo to talk about two new skills designed to keep project knowledge bases sharp. Jess showed how DeX, our internal HR chatbot, went from idea to live in two weeks of hackathon work.

The vote was close enough to tie first time round, which had never happened before. A ten-second timer settled it. Matt’s QA story took the prize for capturing the kind of moment most of us recognise: a real problem, a tight window, and a creative use of AI that unblocked things and surfaced more than anyone expected.

Our recent appointment of Jocelyn Bull as Director of AI Collaboration demonstrates our intent as MSQ DX to curiously explore how AI can enhance the way we work, collaborate and deliver for our clients. Meet Share Question is just one part of that picture.

Same time next month.