New Playwright, LinkedIn's AI Tester, AI for Selenium and More TGNS189

Are you sitting on a Playwright upgrade that finally makes password-less login testable? How do you make your Selenium test into AI-first automation? And have you heard how LinkedIn built a digital tester that found over 200 real bugs on its own? If not, find out on this episode of the Test Guild News Show for the week of June 29th.

Let's grab your favorite cup of coffee and tea and let's do this. All right, first one is all about Playwright. And this is how Pramod wrote all about how 1.61 of Playwright quietly made something testable that blocked his QA teams for over two years, and that's passwordless login.

Andy goes over how this release adds a virtual web auth authenticator through a new credentials API. So your test can register pass keys and answer the browser's credentials ceremonies without any real hardware key. And it works across both Chromium, Firefox, and WebKit. Andy says if you test fintech or banking applications, this alone justifies the upgrade if you haven't upgraded to the latest version of Playwright.

The release also adds a first-class web storage API that retires a pile of page evaluate workarounds, plus matching video retention modes, soft polling assertions, and WebSocket captures and traces. And if you haven't upgraded yet, definitely check out one reason why using those links down below. If your team runs Postman Collections through Newman, there's a new tool you should know about. Kishan tagged me on LinkedIn about an announcement of the release of the Newman Dashboard Orchestrator on NPM, which is a lightweight CLI companion built to pull the run data out of dense terminal logs and into an interactive dashboard.

Here's what it does. It plots past versus failed assertions against API response latency using Chats.js widgets and keeps a local execution ledger tracking up to 50 historical runs on your machine. And there's also a built-in local AI QA advisor, which he says is completely free in privacy first.

It uses a heuristic engine to flag specific faults like authentication expires, server-side 500 crashes, schema regressions, and endpoints that exceed performance budgets. A drill down panel breaks down requests, status codes, and individual assertions. And to run it, all you need to do is install the package globally, point it to your collection, and it spins up local express server on port 8086 in your browser. And if you're testing a public-facing chatbot, I found another thing on LinkedIn by Rohit, who has laid out a framework worth your attention.

He created a post and opens with, on February 2024, a Canadian tribunal ordered Air Canada to pay a bunch of money because the chatbot invented a bereavement fair policy that didn exist His point functional QA is not liability testing So what do you do Well he breaks down seven layers he says to test before launch. First one is functional correctness. Consistency is first since LLMs are mostly correct, isn't a guarantee. Integrity and adversarial safety, a layer that failed on the Air Canada and $1 Chevy Tahoe cases that he writes about as well.

Edge cases like empty messages or a 10,000 character wall of text, groundness for RAG agents, regression, and latency, and costs under load. He names specific tools for each, including prompt food for red teaming, langsmith for turning production transcripts into test cases, and ragas for RAG faithfulness scoring. Also on LinkedIn, I found another post, this one by Jason, all about how to make your Selenium test investment to go AI first. His argument is that the script themselves aren't the asset, but the knowledge trapped inside them is, meaning the business workflows, validation rules, and domain expertise built up over years.

I'm surprised this isn't getting more love because he lays out a four-step transformation. First, he goes over how AI analyzes your suite for readiness, servicing hidden assumptions like environment variables, pre-seated databases, and test sequencing. Second, AI generates questions to recover intent the code never documented. Third, it produces AI-first test definitions that describe outcomes rather than every click.

using what he called OpenTest.ai schema with thinking field that captures reasoning. Quoth is a computer use agent which executes those definitions deciding how to interact at runtime instead of ahead of time. Jason also says this makes tests more resilient and shifts validation from exact string matching to broader checks like whether a page looks visually broken.

I also came across a new tool I haven't heard of before or I may have heard of but I didn't know what it was. This is Momentic which is out with a major platform update which is brand new which is a free access for every developer tester. In this announcement, their CEO frames it around a gap his team kept seeing. AI coding tools let teams ship more code than ever, but how that code gets validated hasn't kept up.

And he points to an AI 2026 report saying monthly incidences are up nearly 58% since AI adoption accelerated. And on May 2026, CloudBees study where 81% of enterprise tech leaders report more production issues tied to AI-generated code. Here what new for testers Every agent runs on a shared knowledge base that adjusts Jira GitHub support docs and Slack so it knows what changing and whether a failure is a real bug or an intended change An Explorer agent reads pull request diffs, finds coverage gaps, and proposes tests without manual authoring.

And they also have a failure classification agent, which triages every failure to a root cause and sends a high signal alert when it's a real bug. Also, the new test format is in tech base written in plain English and executes live against a browser or emulator. And according to them, they said in a few weeks of beta, its agents analyzed over 70,000 test failures, opening 400 pull requests and wrote 600 tests with a 73% merger. So this is a trend I've been seeing.

There's a bunch of tools coming out that do this. So if you haven't checked them out, definitely check them out. Let me know your thoughts in the comments down below. Also, I don't know how many of you know, but I've been hosting the Robocon online event for many, many years.

So I was excited to see this next announcement on LinkedIn. And it's all about a new way that robot framework users can get certification. Tess Smith has announced that its robot framework certification program, Self-Pace Training, is now available, which the describes as the first accredited training platform where you can learn the robot framework skills needed to pass the robot certification exam on your own schedule. And according to Tess Smith, the Self-Pace option joins the existing in-person and online training with the goal of making RFCP certification training more accessible.

So if you wanted the credentials but couldn't commit to a fixed class time, this format lets you work through the material at your own pace. And the course is now live on the Testsmith online learning site. For anyone preparing for the Robot Framework Certification Professional Exam, you can check it out using the link down below. Funny enough, on LinkedIn, I found an article how LinkedIn actually does autonomous testing.

Let's check it out. And this is all about how LinkedIn has gone public with an autonomous testing system it calls QA Agent. And the numbers are really worth looking at. The company says the agent has already found over 200 valid bugs and caught regression in revenue impact flows across iOS, Android, and web applications.

Andy points out some things that if you're a tester, you should care about. Instead of rigid selectors and hidden IDEs, QA Agent uses vision language models to read the screen the way a real person does. It pairs it with a hybrid setup, a fast deterministic replay mode for stable paths, and a slow vision-based planning mode when the UI changes. Several evaluations run as Godreels to filter false positives before any bug gets filed because LinkedIn is optimized for precision over recall And it currently runs 350 plus tests on a 30 cadence and non can now author tests just by using the app in natural language Next up is another real world article.

I found this on Richard Bishop's page on LinkedIn where it talks about how Lloyd Banking Group has shared how it's built an Athena AI-powered knowledge platform that now supports over 35,000 colleagues daily. So this is an example of how to use retrieval augmented generation to ground answers and approved internal content. And the company says average search time fell from 59 seconds to around 20 seconds. For testers, I think the interesting part is how Lloyd keeps the system reliable.

When language models change, outputs have to be tested for faithfulness and relevance. And the group built an internal application called Pegasus to automate that validation. The company also says that that task would have taken subject matter experts over 160 hours of manual review, where Pegasus completed in just over an hour. And they're saying Athena is now becoming a shared platform with API access for other tools and AI agents.

Like I said, I like real world examples, how to implement AI and how to automate things. And here's one that you should definitely check out as well. Last up is all about security. And I found it on a friend of the show, Iran Kinsbruner.

So Iran helped contribute to this report by Jamie Hampton, all about how OpenAI has announced an expansion of its Daybreak cybersecurity program. And its focus this time is on remediation. So the update includes an updated codex security plugin, a more capable version of GPT 5.5 Cyber, a partner program and an open source initiative.

And if you're a tester, this is why you should pay attention because the Codex security plugin can scan a whole code base, a portion or a single commit, then identifies a flaw of severity and location, trace attack paths and proposes a patch. It can also ingest findings from scanners, bug bounty reports and ticketing systems, validate them and then export results in different formats. And they say since March, OpenAI says it's scanning over 30 million commits across more than 30,000 code bases.

But this article does come with some pushback. One of them is from HackerOne's Needle, who says validation will become the defining bottleneck and checkmarks of Ron Kinsbruner, one that you can't ask the same class of model that writes the code to be the sole judge of its security. All right, for links of everything to value we covered in this news episode, head on over to all those links in the comment down below. So that's it for this episode of the Test Guild News Show.

I'm Joe. My mission is to help you succeed in creating end-to-end full-stack pipeline automation awesomeness. And as always, test everything to keep the good. Cheers.