Vibium, Autonomous Testing, and the CORS Model for AI Agent Security

Did Jason Huggins just make selectors obsolete? What major test platform just rebranded around autonomous AI agents? And what if today's AI agents are repeating the same exact security mistakes the web made decades ago? Well, Mindo, in this episode of the Tuskill News show for the week of January 12, 2026. So grab your favorite cup of coffee and tea and let's do this. All right, after a month away, I'm back and ready to dive into what you all have missed. While I was on vacation, the biggest news to drop during my time off was the official release of Vobium by Jason Huggins, the original creator of both Selenium and Appium. Jason has now launched his third major testing framework and it's now available for you on GitHub. So finally, after all the hype Jason has been building up after months on LinkedIn Vim is now live and available for anyone to check out and explore and see for themselves whether you're ready to or not in the post select error generation Jason has been talking about all along for the past few months as I mentioned on LinkedIn. So if you haven't already, I highly recommend you check it out using that link down below. And speaking of volume, I actually included it in my yearly Automation Testing Trends report, which I just released as well. So the biggest surprise based on all the data I collected in 2025 is that 72.8% of testers say that AI powered testing is their top priority. Yet the most upvoted question from all our webinars was whether AI actually makes testing obsolete or more critical now than ever. And some other findings I go over is the growing integration crisis as Playwright surges past Selenium, but teams can't seem to make their stack work together. I also cover a massive AI skill gap even among senior testers, Vibe coding and Vibe testing creating new security and quality risk, a shift away from theory towards real voice working AI testing Examples why accessibility laws in 2025 will suddenly matter to you and every global team you're on, and how agentic AI MCP servers and tools like VIBM could make traditional selectors obsolete. I also share a practical action plan for 2026, what to pilot, what to ignore, and how to avoid betting your testing strategy on hype. And if you really want to know where AI testing is actually heading, make sure to check out my full article down below. And also register for Automation Guild, where we cover a lot of these topics that people are struggling with, so you definitely want to check that out as well. So speaking of practical examples you might want to know how do humans actually collaborate with AI? Well, I found an article just for you. This is by Jeff Morgan, who shared details of the Stride workflow. It's a system designed to let humans and AI collaborate without forcing full autonomy. It gives developers control over when and how AI agents step in, especially for teams not ready to fully hand off work. And he goes over how Stride has three human intervention points before work starts, humans review and adjust task and add context. Second, during the workflow configuration, hooks run actions like pull code, a running test and security checks in the final one is before completion, tasks get flagged, needs review so humans approve or request changes before an AI agent finishes. So the work flows through backlog ready doing review and done. And agents pull tasks from ready remove, run pre and post execution hooks and send flag tasks to review or move them straight to done. Approved tasks trigger final hooks before closing and he also gives an example of how hooks can run git pulls, test coverage, formatting, credo and Sablo security scans and because agents work much faster than humans, Jeff recommends keeping far more tasks in Ready than you would in normal dev processes, and his task assignment is based on five rules Ready status, agent capability, match complete dependencies, file conflict avoidance, and priority with positions as a tiebreaker here's another announcement that came my way via my LinkedIn feed and this by John Ferguson Smart, who's announced that Serenity BDD now supports playwright in version 5.10 and the integration works by allowing users to register their Playwright page with Serenity, which then automatically captures screenshots at each step method. Testers can now continue using Playwright's full capabilities while gaining access to Serenity's BDD reporting features. This is one of the top tools I ever used when I was a full time automation engineer. I really love Serenity, love how they're embracing Playwright. Also, John goes on to describe this as an early release and requesting feedback from the Playwright community about what would make the integration more valuable for the workflow. All right, another surprise that caught my attention already in 2026 is a major rebrand for a huge testing company and this is how LambdaTest has rebranded to test Mu AI and re architect its platform to be AI native, deploying autonomous AI agents for planning, authoring and orchestrating and analyzing software quality with minimal manual intervention. And asad, the CEO and co founder, states that with billions of tests running on their platform are now delivering experiences where human ingenuity and machine intelligence combine to make quality engineering effortlessly powerful. The platform now offers two core components, autonomous AI agents for testing that plan, author and evolve end to end tests using company wide context or natural language prompts covering databases, APIs, UI performance tests and more and an agency AI test cloud that provides scalable and unified test execution for visual regression accessibility, API performance testing, web and mobile and custom enterprise environments and the Testament name was adopted directly from the company's community. Since 2022, the Testament Conference, which I've had the honor of opening every year, has served as the industry's primary forum for advancements in AI and quality engineering. And congratulations everyone now at testmu AI for your huge growth in all you've done for the industry over the past few years. All right, if you do anything with API automation, this next article is how Nitesh has announced the completion of an AI automation project he's created using Postman, and the project demonstrates several technical implementations including Capturing and importing APIs into postman collections utilizing postman's variable types including local collection environment, global data and random variables and executing collections through Newman, which is Described as a JavaScript command line interface and if you check it out on GitHub, the project also includes report deployments to GitHub pages and email notification configured through GitHub Actions, and the work covers API fundamentals including request components and distinctions between authentication, authorization and testing for functionality, schema validations and data integration. If you're looking to get your hands dirty on a project to learn more about API testing, this is something you should definitely check out. Let me know your thoughts. And another example of automation in the real world is another article I found on LinkedIn. This is how Marwan has announced completion of an AI powered playwright automation framework for the US.gov that integrates AI agents and large language models. According to him, the framework includes multi browser testing across chromium, Firefox and WebKit security first CICD with AI enhanced vulnerability detection, WCAG accessibility compliance with intelligent edge case generation as well as smart parallel execution delivering feedback in under five minutes, according to this post. And what I thought was cool is he says the AI and LLM innovation features include AI test agents for autonomous test creation and maintenance using large language models intelligent test selection with LLM powered analysis identifies critical test paths based on code changes, self healing tests where AI automatically detects and repairs broken selectors and flaky tests, natural language test creation that converts plain English requirements into executable playwright tasks and smart test data generation with context aware test data that is created by LLMs that understand federal data requirements. There's a lot there and you can find it on GitHub. He also describes the framework as a production ready rather than a proof of concept. So you definitely should see if you give it a spin and emphasizes that federal systems demand reliability and security, so you know that something like this is probably something you should check out. And he states the framework combines traditional automation with modern AI intelligence. So another awesome real world project you definitely should check out for yourself using that link down below and last up is a security news item. Clinton and Alex from Block proposed a core style security model for AI agents to stop context injection attacks. Their idea treats the LLM as untested code and an agent as a browser and MCP tools as servers, fixing the fact that agents can't reliably separate data from instructions. And they go over a common attack where poison content like an email can inject commands that make an agent call unauthorized tools. Block solution is to track which tools were called. Since the last human input of a new tool call appears after another tool ran, it's treated like a cross origin request. It must be explicitly authorized or it fails. They also suggest removing tool responses from the content window between user turns to limit inter turn manipulation and even if that means rerunning some tools. And they acknowledge that this approach has limits, it assumes a trust agent and doesn't stop all second order prompt attacks. But Block is building a proof of concept for the goose agent and inviting feedback on this approach on GitHub as well. Alright flicks of heritage 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 help you succeed in creating end to end full stack pipeline automation awesomeness. As always, test everything and keep the good cheers.