Welcome to the 328th issue!

There's something I've been waiting for.

Google has open-sourced Artemis — an AI agent that tests Android apps on real phones and emulators.

However, Nicolas Dehandschoewercker claims that Artemis reused code from their open-source mobile-use project without credit — I expected better from Google.

Luckily, a day after that post, Google added the missing attribution.

Happy testing!

Dawid Dylowicz

It's always interesting to see how experienced testers look back at their career and share what they've learned. Vishal Gaikar did just that after nearly two decades.

Moreover, Ramella Basenko adds one more lesson and explains why The maturity of your QA process is capped by the maturity of your handoffs.

Wondering why your AI-generated tests don't find edge cases? Amruta Lohabare shares the prompt they now use to catch what it missed.

Also, for a wider view, Jason Arbon compares Testing: Human Experts vs. AI.

Rafaela Azevedo found an interesting trend — tech job openings now ask for more years of experience and list fewer skills. This signals that good judgment and experience matter more than pure tooling and coding excellence.

At the same time, Jaclyn Leigh looks beyond automation in A Senior QE's Take on Modern Quality Engineering.

Tired of testing everything at the end of a sprint? Julia Pottinger shows how doing refinements and creating a shared Definition of Done can help you shift left.

Similarly, Matthew Sullivan asks what's left of agile in The Agile Zombie, while Khurram Bilal suggests Three Parallel Workstreams: Design, Build & Test on Day One.

Manual QA is dead? The contrary seems to be true. Someone on Reddit hired a junior tester whose thoroughness changed their mind and the community agrees.

In a related thread, someone asks: QA folks, how are you feeling about things these days?, to which Betty Lin responds Why I Am Still Doing QA.

We ship code faster, but we don't test faster. Devjyoti Das suggests that we should stop acting as a gate and price the risk of each change — just like insurance companies.

Also, Keith Klain says that Faster Agents Don't Remove the Bottleneck. They Move It, which may lead to Death by a Thousand Prompts, as Zyane Tan points out.

AI can help a tester write 30+ tests a day. But someone still has to understand and own that code. Vaibhav Singhal shares simple habits to keep it easy to maintain.

Similarly, Amruta Lohabare advises us to Stop Using AI Like a Fancy Autocomplete. Your Test Suite Deserves More.

Does your dashboard report 90% automation? Ravi Rawal shows why, even at that number, you can miss important failures, like a lost network connection or an expired token.

On that note, Balasundar Veluchamy asks: Your API Tests Pass. How Deeply?

An interesting read from Tristan Rhodes outlining how to make computer-use AI agents work as a QA engineer. It tests each new feature overnight and reports bugs with screenshots and reproduction steps.

Furthermore, if you plan to build your own, Swati Seela shares: Can an AI Agent Perform Exploratory Testing? 5 Lessons from Building One.

Internal test frameworks may grow over time without much planning. Muskan Agarwal says it's risky and explains why we should treat it as a public library, with versions, a changelog and working examples.

Also, Muskan says that Every Bug Report Is a Test Case You Forgot to Write.

Running LLM evals in CI? Pratham Mhatre shares a story about how a single run may not always provide enough evidence, and how to properly interpret the results.

What's more, Jaren Charles Cudilla describes Database QA: The Bugs Your UI Tests Will Never Catch.

If you're struggling to debug a flaky Playwright test, Anton Gulin advises keeping a trace of the failed attempt, not just the retry that passed.

Also, Richard Forjoe shares another useful Playwright tip in Playwright — Run Only the E2E Tests a Pull Request Can Actually Break.

Playwright 1.63 is a significant release and Fidaa Berrjeb describes the new features with code examples: test locks, locator.visible() and clearer steps in the trace viewer that can make your test suite easier to run and debug.

Similarly, Svetlana Tretjakova gives another good overview in Playwright 1.63 Is Here: The Updates QA Engineers Should Actually Care About.

Thinking about AI-generated Playwright tests? Michael Brown compares Playwright MCP with codegen and explains when to use each one, including advice on security.

Moreover, Svetlana Tretjakova has good advice for any Playwright suite: Stop Mocking Everything: Advanced Network Interception in Playwright

If you write Node.js and TypeScript APIs, Chirag Mehta has a hands-on guide to testing them with Vitest and Supertest, including how to mock and gate CI on coverage.

And if you use Java, here's a good question by Andrii Olieinik: What Is a Test Fixture in JUnit, and Why Is @BeforeEach Not the Fixture?

An unusual book review from Mike Harris on what we can learn from the team behind the Sidewinder missile, like fast feedback from testing and close contact with the people who use the product.

Continuing the series of interviews, this time Joe Colantonio invited Jonathon Wright to discuss the impact of agents on testers' role and test automation in this 34-minute recording.

I mentioned above the new features in Playwright 1.63, and here's a more detailed deep dive by Joan Esquivel Montero into Test Locks.

In addition to that, Jérémie Fleurant explains what's Beyond Playwright Test Locks: Pooling Shared Test Users.

Must be an AI issue... 😂

Thanks for reading!

If you enjoy this newsletter and find it helpful in becoming a better tester, please consider sharing it with others.

Dawid Dylowicz