Automation Debt, AI Validation Crisis & Cypress Trust Gate Launch

Are you spending 80% of your time fixing tests that never catch bugs? What are the two must-know Cypress plugins that are in the news this week? And is AI flooding your codebase with code nobody can validate? Want to know more? Find out in this episode of the TestGuild News Show for the week of March 15th. So grab your favorite cup of coffee or tea and let's do this. And before we get into it, if you haven't already, please make sure to subscribe to our TestGuild LinkedIn News Show newsletter. I'll have a link for it down below to never miss another episode. First up is a problem that testers have been dealing with forever: test maintenance. So this article, David Burns, who is a well-known Selenium contributor and engineering leader, says many test automation suites eventually turn into trust problems when flaky tests and unreliable results pile up. And he cites research showing teams could spend up to 80% of their time maintaining tests instead of building new features, a cycle he calls maintenance theater, where teams stay busy fixing scripts without improving the actual product quality. So Dave proposes a simple framework called Keep, Refactor, or Kill. Keep, stabilize high-value tests that cover critical user journeys. Refactor, flaky tests that are reliable but poorly written. And Kill, redundant tests, ghost tests, or removed features, or tests that haven't caught a bug in years. He also highlights the AI agent paradox. AI tools can generate hundreds of tests in seconds, but they often miss real business risk. An AI might create dozens of tests checking a button's color while ignoring a race condition in checkout logic. And without human oversight, this will lead to test explosions, misleading green dashboards, and thousands of passing tests that don't validate real software behavior. So David argues that teams shift from measuring code coverage to measuring risk coverage, focusing tests on failures that would actually hurt the business. And he calls this approach confidence engineering. Use AI to help generate tests, but let humans guide it towards the most critical logic and failure modes. And he also offers a quick exercise for you to try right now. Open your test suite, find the test that fails most often, and if it's not protecting a mission-critical flow, delete it to immediately improve pipeline reliability. Great advice as always from David. Highly recommend you give him a follow and check this out more using the link down below. Next up is our webinar of the week. So TestGuild is hosting a webinar on March 19th featuring Don Jackson, who is a technical evangelist at Perforce. If you don't know him, he's an awesome speaker, and he's going to be presenting an in-depth review of the State of DevOps Report: AI and Testing Edition 2026. The webinar is going to focus on governance, cost maintenance, and continuous quality in software testing, and the session addresses challenges facing teams with bottlenecks in automation, fragmented toolchains, and scaling DevOps practices. And Don promises he's going to provide awesome industry benchmarks and best practices to helping you maintain a competitive advantage. So I highly recommend you join to gain insights on market trends, redefining testing, including the role of AI in governance and cost management. The session will cover practical steps in streamlining automation, benchmark performance, and help sharpen your competitive edge. So register now using that link down below to discover AI-powered testing insights for measurable business value. You don't want to miss it. Check it out. Hope to see you there. Next up is all about a Cypress accessibility plugin you should be using by now. And this is Sebastian announced a new release of WIC-A11y 3.01 has been released, helping bring compatibility with Cypress 15.10 and upcoming Cypress 16 without requiring users to change their existing configs. And according to the post, this update addresses a major Cypress change, the deprecation of cypress.environment. For security reasons, Cypress now separates environment variables into two types, cy.env for private values and cypress.env exposed for non-sensitive values. This normally would force users to reconstruct their configs or CLI scripts, but instead what he's done with this plugin, it now supports both exposed and non-exposed variables. Also on my LinkedIn feed, I came across a new open source tool called Cypress TrustGate, and this aims to catch hidden quality issues in Cypress test suites before they slow down delivery. And this was created by Saran Kumar, and the tool targets 4 common problems that quickly degrade test reliability. Flaky tests, accessibility violations, slow or poorly written tests, and fragile selectors. And if you go to GitHub, you can read more, but it analyzes Cypress suites across multiple dimensions to detect flaky tests and runs test suites repeatedly to measure stability. It also integrates with axe-core to automatically scan for accessibility violations during test execution. On top of that, it performs static analysis to flag slow tests and anti-patterns like hard-coded waits or repeated authenticated steps. It also scans test files for brittle selectors such as nth-child, deep CSS chains, and chain class selectors. All the results then are combined into what's called a trust score from 0 to 100. Scores above 90 indicate strong test health, and scores below 35 signal critical issues. The tool generates both JSON outputs and HTML dashboard highlighting per-test stability scores, accessibility issues, and slowest tests and recommended fixes. Looks like a cool open source solution. Cypress users now have a free way to monitor test reliability, accessibility, performance, and selector stability. This sounds like something you're into. Definitely check it out. Let me know how it works for you in the comments down below. Another thing that caught my attention this week was a new report from Harness, which highlights a growing problem in modern software development. AI is speeding up coding, but the rest of the delivery pipeline hasn't caught up. And in the State of DevOps Modernization 2026 report, Serving 700 engineers and technical managers across different countries, the finding shows that while AI tools help developers write code faster, testing, security, and deployment systems often struggle to keep up. Teams using AI coding tools multiple times per day deploy more frequently, but they also report more issues. So 69% of heavy AI users say deployments involved AI-generated code frequently cause problems. When incidents occur, these teams take an average of 7.6 hours to resolve them. Compared with 6.3 hours for occasional AI users. AI adoption is also increasing manual work downstream with 47% of frequent AI users saying tasks like AI validations, remediation, and release checks have become more difficult. The report also finds that developers still spend about 36% of their time on repetitive manual work. A major reason, most organizations lack standardized delivery systems. And really, two of the biggest numbers that really popped up to me is 51% see an increase in code quality problems and 53% report more vulnerabilities or security incidences. And that harness data we just went over makes a lot more sense when you understand the deeper problem, which luckily Philip has just published a piece that reframes the entire AI coding conversation. So in this post, "Is Code Evolving or Mutating?" Philip argues that AI coding tools are changing the core bottlenecks in software development. The core challenge is no longer writing code, it's deciding which AI-generated code should actually ship to production. For example, he points to Anthropic's new Claude code review feature where AI agents automatically review pull requests for bugs. When critics questioned AI reviewing its own code, Claude code's creator Boris responded that the solution is more AI agents reviewing each other using additional tokens to catch mistakes introduced by other agents. And Philip goes on to highlight a comparison he's seen from others comparing AI coding to evolutionary biology. So in this model, mutation equals generating new code, selection equals choosing which code survives, and persistent equals keeping the working solutions. AI dramatically increases the mutation rate. Agents can generate features, refactor codebase, and propose architectural changes quickly, but the selection layer hasn't scaled at the same pace, creating a new bottleneck. And that's where quality engineers come in. Philip argues that testing and verification now acts as the selection mechanism. If testing doesn't keep up with the flood of AI-generated changes, systems won't improve and they'll simply just drift into instability. A lot more good stuff from Philip you could definitely check out using the link down below. And if you're doing anything with Playwright, I found the new post TestDino written by Daruv that breaks down the emerging Playwright AI ecosystem and how it aims to solve one of test automation's biggest headaches, test breaking every time the UI changes. The system combines model context protocol, built-in AI agents, and accessibility tree-based automation to make tests more resilient. So instead of guessing selectors from screenshots, the AI uses defined browser actions like navigate, click, type, and snapshot. Second is the agent layer, which introduces 3 AI agents: the planner, the generator, and the healer. Microsoft benchmarks cited in the article that shows the healer fixed over 75% of selector-related failures. And the key difference from many AI testing tools is that Playwright relies on the browser's accessibility tree, not screenshots. So the accessibility tree exposes structured information about every element on the page, roles, name, state, and hierarchy. And this approach is faster, more deterministic, and less sensitive to visual style changes. And the article also maintains that most teams won't see instant automation gains. The biggest payoff comes from reducing test maintenance, especially for teams spending significant time fixing broken selectors. All right, for links of everything of value we covered in this news episode, head on over to the links in the comment down below. So that's it for this episode of the TestGuild News Show. I'm JoeBot. My mission is to help you succeed in creating end-to-end full-stack pipeline automation awesomeness. And as always, test everything and keep the good. Cheers.