4 October 2026 Oussama Belakhdar 3 min read

Case study: from a PDF specification to Playwright tests, three weeks became one day

A test automation engagement for a software vendor, in 2026. The starting point: a functional specification in PDF, to be turned into test scenarios and then into automated tests with Playwright. Here is what an AI assistant changed, and the conditions for the gain to hold.

Key takeaways

About three weeks cut to one day, from the PDF specification to Playwright tests.
The gain comes from two steps: extracting the scenarios and writing the test code.
It only holds with systematic review: AI can invent a business rule.
One rule above all: a client specification only goes into an AI tool with the client's consent.

The context

A software vendor, a functional specification delivered as a PDF, and a classic goal: derive the test scenarios from it, then automate the ones that must run on every release. We do not name the client.

Before and after

Step	Before	With an AI assistant
Extract the test scenarios from the PDF specification	about 1 week	half a day
Automate them with Playwright	about 2 weeks	half a day
Total	about 3 weeks	about 1 day

These durations are the ones observed on this engagement. They depend on the size and quality of the specification: this is not a promise for every project.

The conditions for the gain to hold

Review every generated scenario. An assistant can propose a case based on a rule that is not in the specification. Each scenario must be traced to a specific requirement.
Review the test code. We use our review checklist for AI-generated tests: business assertion, accessible locators, isolated data, secrets kept out of prompts, never on production, limited exploration, test seen failing at least once.
Get the client's consent before sending a specification to an AI tool, and only use a tool the client authorises.

What AI did not change

Deciding what deserves to be tested, and in what order.
Judging the remaining risk before a release.
Explaining to the team why a test fails, and what it protects.

And for your team?

If your test campaigns still take long to prepare, or if your team is starting to use AI assistants without a review method, a QA audit gives you a clear picture: a written assessment, a test strategy and a costed action plan, for €1,000 excl. VAT as a fixed fee. The first scoping session is free. See also software testing by industry.

Frequently asked questions

How much time did AI save on this engagement?

Starting from a PDF specification, extracting the test scenarios went from about one week to half a day, and automating them with Playwright from about two weeks to half a day: about three weeks cut to one day.

Can a client specification be sent to an AI assistant?

Only with the client's consent, and in a tool the client authorises. A specification can contain business rules or confidential data.

Are AI-generated tests reliable?

Not without review. AI can invent a business rule that is not in the specification, or write a test that checks a click rather than a business result. Every scenario and every test must be reviewed, and every test seen failing at least once.

How much does a QA audit cost at AutomationDataCamp?

The QA audit and diagnosis costs €1,000 excl. VAT as a fixed fee: a written assessment, a test strategy and a costed action plan. The first scoping session is free.

Engagement report by Oussama Belakhdar (2026), client anonymised. Written with the help of AI; the durations are those reported by the author for this engagement.