Welcome to the 332nd issue!

Atlassian just published The State of Developer Experience in 2026.

It's a survey of 1,500+ developers and engineering leaders. Most of it is about AI strategy but a few numbers stood out to me:

• 73% say AI-driven speed can come at a cost to quality,
• 65% spend time correcting AI-generated work that missed the original intent,
• and the biggest shift in the dev role: 41% now have more responsibility for validating the quality and safety of AI output.

And in separate data from 500+ companies, confidence in changes dropped by 6% in just one quarter.

So developers do more verification and testing, but they trust the code they push to production less.

Interesting times.

Happy testing! 🙂

Dawid Dylowicz

Want to do spec-driven development with AI agents? Masoud Bahrami has some tips on how to approach it the right way, so that it doesn't drift too much.

Additionally, Josphine Job tells us more on Why QA Experience Is an Unexpected Advantage in Spec-Driven Development, while Gal Arav continues the topic of Towards Spec-Driven Test Automation: Part 2.

Anılcan Gül shares how they merged tester and developer roles once AI agents made testing the bottleneck. Developers now test their own changes, with pair testing for the risky ones.

Also, Jean-François Marquis has a similar story from another company: From Tester to Trust Arbiter: Our Experience with QA in the Agentic Engineering Era at ADEO.

A good reminder from Jeff Nyman on how AI-assisted development creates comprehension and cognitive debt, and how testing shows where our confidence isn't backed by evidence.

At the same time, Vineet Verma points out that AI can write tests from a user story. It can't tell you the story is wrong, while Bob Chen has a good reminder that The Requirement Is Not the Ticket. It's Everything Around It.

This is useful. You know that feeling whenever the pipeline fails and devs ask you to debug it? Mohamad Andika Bhaskara shares how their internal LLM now sorts each failure in the pipeline.

Similarly, Julia Pottinger gives a practical example of From Bug Report to Beta Build: A QA Release Workflow.

It actually may not be the right question. Michael Brown splits testers' work into four zones and shows which tasks AI already does and which need a tester.

Additionally, Ömer Yusuf gives a similar view on why AI Won't Replace Your Testing Mindset: How QA Engineers Should Use AI.

Test automation doesn't remove the bottleneck but it moves it to reading the results. Bavo Bruylandt describes how their AI agent ranks the failures overnight, followed by the second part about Building Agentic Test Triage.

Someone on Reddit asked the testing community for advice on implementing AI-driven automation and, (un)surprisingly, most people agree that AI coding assistants helped them achieve better results.

However, Niklas Gruhn from Anthropic warns about Atrocious AI-written tests.

If you want to let an AI agent push code or approve pull requests, you'll want to read this. Marina Polkovnikova gives practical ways to test the hooks that block it, including tests for prompt injection.

Furthermore, Amruta Lohabare reminds us that AI Tests a Moment. Your Users Live in Time.

Kochi Chuang describes how to review tests written by an AI agent, suggesting mutation testing and a checklist of ten questions instead, followed up by the second part.

Speaking of that, Mariya Ushakova asks: Your AI-Generated Test Passes. Who Decided the Expected Result?

Checking the code that agents write is now the new bottleneck. Addy Osmani⁠ describes how E2E, property-based and differential tests let the agent check its own work before we review it.

In relation to that, Tim Lolkema tells us more on Why you should start using agentic testing.

Want to test Android and iOS apps with Maestro? Ravi Rawal shares how to keep the test suite stable and easy to debug as it grows.

Thinking about letting an AI agent write your API tests? Mohd Aatif Khan gives the rules they had to teach the agents, such as one ID variable per resource and 204 for void endpoints.

Chinmay Jha shares Optics — an open-source framework where you write tests as CSV or YAML rows and run them on Android, iOS, web and TV.

You can find more practical details in the following two articles:

• Your AI agent writes the test once. Optics runs it with no model.
• Your tests passed. Your app didn't.

Since web pages can now give AI agents their own set of tools to interact via MCP, Svetlana Tretjakova shares why that needs its own tests and how to write them in Playwright.

Moreover, Vitaliy Potapov describes a way of Testing browser extensions with playwright-webext.

In this 14-minute recording, Kacper Ciepielewski brings up several good arguments on why we should be concerned about vibe testing with AI.

Acceptance criteria... 😂

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