Welcome to the 334th issue!

I found this read from Anthropic on formal software verification:

LLMs are making formal software verification feasible at a much larger scale

It matches what I've been thinking lately about testing in general.

We avoided big test suites because they were flaky and hard to maintain, and that was the right call. Exhaustive testing was simply too expensive.

AI is changing that. It's not perfect, but with significantly cheaper maintenance, we can add more checks on top.

But more checks only help if they verify the right thing. That's still our job.

Happy testing! 🙂

Dawid Dylowicz

Jitesh Gosai describes how to control what agents do and suggests we start small and learn what works instead of trusting them by default when they say they've finished.

Moreover, Rachel Laycock wonders: Why are good developers so wary of agentic development?

We often check what AI writes for us. But what if AI reviews our work? Ramella Basenko shares a story from a project and suggests what guardrails we should implement.

Moreover, Josphine Job wrote two good articles on What Happens to Your QA Brain Every Time You Reach for AI First and Why My Agentic Test Pipeline Is Not Exploratory Testing.

A tester on Reddit helps build an AI agent that may soon do most of their job. Others in the thread share what happened when their company tried the same thing.

And there are more related conversations:

• Any other QA folks laid off because of AI?
• For people who hire QA, what on a resume makes you stop reading

Richard van de Laarschot shows why verification is the new bottleneck and how a smaller, clearer test suite can help with agentic development.

Similarly, Vineet Verma looks at how 70% of Testers Use AI to Write Test Cases. Only 20% Use It to Find Risk.

Most of us try to improve our testing process from our own desk. Marina Jordão suggests talking to developers, product managers and tech leads first, including questions to ask each group.

At the same time, Alex Stetcenko describes Several teams, one small test automation group: why quality is a team sport.

Thinking about contract testing? Bas Dijkstra shares what often goes wrong when you start with contract testing, from picking Pact too early to dropping end-to-end tests.

Additionally, Gal Arav continues the series with Towards Spec-Driven Test Automation: Part 3.

Many of us have a big UI suite full of failing tests. Kavitha Namduri explains how to use an AI agent to group failures by root cause and fix them while maintaining coverage.

On that note, Indra Aristya describes what happens When AI Makes Automation Suite Harder to Maintain.

An interesting experiment by Ujjwal Kumar Singh trying to see how effective an agent can be in exploratory testing. It found single-input bugs easily but almost never found timing bugs. You can also catch up on part 1.

Furthermore, Amr Salem outlines how to get from AI Agent Anatomy to Reliable QA Use Cases.

A good test ratio may not tell you whether you test the right things. John Ringler describes how to pick each test level by risk, followed by an explanation of why The Hardest Part of Test Automation Is Deciding What Not to Automate.

A helpful article from Anish Naik and Marcelo Morales on using two AI agent plugins: one for fixing reported bugs and the other for checking if the fix introduced any new regression.

In relation to that, Anton Gulin asks: What should an AI code review read?

Does your test suite pass only in one order? Svetlana Tretjakova describes how to find hidden dependencies with a --shuffle option in Playwright and fix them.

Moreover, Saqib Jan wrote about locators in So Enterprise Playwright Suites Are Running the Locator Hierarchy in Reverse.

Lauren Leidal compares a coding harness with a testing harness for AI agents and says why the same model finds more bugs in the second one. You can also read the second part on Scaling testing agents across a company.

Ben Sabic from Vercel has just announced the agent-desktop — a new solution that controls desktop apps on macOS, Windows and Linux. This guide explains when to use it compared with the previously released agent-browser.

Struggling with slow end-to-end tests? John Ringler shows where the time goes in Cypress and Playwright runs and how to cut execution time.

Also, Jennifer Shehane gives advice on How to choose the right Cypress plugins for your project.

Si++... 😂

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