<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Ashwini Todewar on Medium]]></title>
        <description><![CDATA[Stories by Ashwini Todewar on Medium]]></description>
        <link>https://medium.com/@ashwini.todewar?source=rss-cbcae1e345c4------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*dmbNkD5D-u45r44go_cf0g.png</url>
            <title>Stories by Ashwini Todewar on Medium</title>
            <link>https://medium.com/@ashwini.todewar?source=rss-cbcae1e345c4------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sun, 22 Nov 2026 05:17:43 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@ashwini.todewar/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Automating Voice Input with Playwright, Edge TTS and a Fake Microphone]]></title>
            <link>https://medium.com/@ashwini.todewar/automating-voice-input-with-playwright-edge-tts-and-a-fake-microphone-3ba323070d49?source=rss-cbcae1e345c4------2</link>
            <guid isPermaLink="false">https://medium.com/p/3ba323070d49</guid>
            <category><![CDATA[voicebots]]></category>
            <category><![CDATA[automation]]></category>
            <category><![CDATA[bots]]></category>
            <dc:creator><![CDATA[Ashwini Todewar]]></dc:creator>
            <pubDate>Wed, 30 Sep 2026 11:06:53 GMT</pubDate>
            <atom:updated>2026-09-30T11:11:23.105Z</atom:updated>
            <content:encoded><![CDATA[<p>Voice automation is slightly different from normal UI automation. With UI tests, we can click a button, enter text, and verify the result. With a voice application, the test needs to provide <strong>actual audio input</strong>.</p><p>So I built a small proof of concept to automate voice input using <strong>Playwright, Edge TTS, FFmpeg, and a fake microphone</strong>.</p><p><strong>Repository:</strong> <a href="https://github.com/ashwinitodewar491/voice-bot-automation">https://github.com/ashwinitodewar491/voice-bot-automation</a></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*fnxKcBhtgFJF1QBCbAXTSg.jpeg" /></figure><p><strong>What I Wanted to Achieve<br></strong>The goal was simple:</p><p><strong>I wanted Playwright to send a voice query to a browser without a human actually speaking.</strong></p><p>Instead of using a physical microphone, the test generates an audio file and makes Chromium treat that file as microphone input.</p><p>The basic flow is:<br><strong>Test Text → Edge TTS → WAV Audio → Fake Microphone → Browser → Speech-to-Text<br></strong>This makes the voice interaction completely automatable.</p><h3>Tech Stack</h3><p>The project uses:</p><ul><li><strong>Java</strong></li><li><strong>Playwright</strong></li><li><strong>TestNG</strong></li><li><strong>Edge TTS</strong></li><li><strong>FFmpeg</strong></li><li><strong>Chromium</strong></li></ul><p>Playwright handles the browser automation, while Edge TTS is responsible for generating the speech.FFmpeg is used to prepare the generated audio in the required format.</p><h3>Generating Voice with Edge TTS</h3><p>I didn’t want to maintain pre-recorded audio files for every test case.Instead, the test starts with text.<br>For example:</p><pre>What is my account balance?</pre><p>Edge TTS converts this text into speech and generates an audio file.</p><p>One reason I chose Edge TTS is that it is <strong>free to use and not tied to a specific operating system</strong>. This makes the voice-generation part easier to run across different environments instead of depending on an OS-specific speech engine.</p><p>It also provides support for a <strong>wide range of Indian languages and voices</strong>, making it useful when testing multilingual voice applications. For example, the automation can generate speech in languages such as Hindi, Bengali, Marathi, Telugu, Kannada, Malayalam, Gujarati, Punjabi, and Urdu, depending on the available voices.</p><p>This makes the test data easy to maintain. If I want to test another voice query or language, I only need to change the input text and voice configuration.</p><h3>Making the Audio Work as a Microphone</h3><p>Generating the audio file is only half of the solution.The browser still expects input from a microphone.For this, Chromium can be configured to use an audio file as the microphone source.The test launches the browser with the required fake-audio configuration and points it to the generated WAV file.</p><p>From the application’s perspective, it receives microphone input normally.There is no physical microphone involved.</p><h3>Why FFmpeg?</h3><p>The TTS-generated audio is not always in the exact format required for reliable browser microphone simulation.</p><p>I use FFmpeg to convert the audio into a suitable WAV format.</p><p>This also gives better control over properties such as:</p><ul><li>Sample rate</li><li>Audio channels</li><li>PCM format</li><li>Silence before and after speech</li></ul><p>That last point is particularly useful.</p><p>I added a small amount of silence before the actual speech so that the beginning of the sentence is not missed when recording starts.</p><h3>Demo Target</h3><p>For the demo, I used <strong>Whisper Web by Xenova</strong>, a free, open-source, in-browser speech-to-text application.</p><p>The test performs a simple flow:</p><p><strong>Open Whisper Web → Start Recording → Send Fake Microphone Audio → Stop Recording → Verify Transcription</strong></p><p>This provides a simple way to demonstrate that the browser is actually receiving the generated voice input.</p><p>One important point: <strong>Whisper Web is a third-party demo target and is not maintained by me.</strong></p><p>Because of that, changes to its UI can break the automation. For example, if a button is renamed or its locator changes, the test may fail.</p><p>That’s expected for a demo project.</p><p>For real regression testing, the same approach can be used against <strong>your own voice-enabled application</strong>, where you control the UI and application behaviour.</p><h3>A Small but Important Challenge</h3><p>Timing is one of the areas that needs attention in voice automation.</p><p>The application needs some time to start recording before the audio begins.</p><p>If the audio starts immediately, the first part of the sentence can be missed.</p><p>Adding a short silence at the beginning of the audio helped make the automation more reliable.</p><p>Similarly, a little silence at the end gives the speech recognition system time to process the complete input.</p><h3>What This POC Demonstrates</h3><p>The main thing I wanted to prove with this project was:</p><p><strong>A voice interaction can be automated without a human speaking into a microphone.<br></strong>The same concept can be extended to a real voice bot.</p><p>For example:</p><p><strong>Login → Start Voice Interaction → Send Voice Query → Wait for Bot Response → Validate Response</strong></p><p>The fake-microphone setup and Edge TTS generation can remain the same while the demo target is replaced with the actual application.</p><h3>Final Thoughts</h3><p>This project was a practical experiment in bringing voice input into browser automation.</p><p>The combination of <strong>Playwright + Edge TTS + FFmpeg + Chromium fake microphone</strong> provides a simple way to generate repeatable voice input for automated tests.</p><p>It removes the need for a human to speak during every execution and makes voice scenarios easier to include in a regular automation pipeline.</p><p>The complete implementation is available here:</p><p><strong>GitHub:</strong> <a href="https://github.com/ashwinitodewar491/voice-bot-automation">https://github.com/ashwinitodewar491/voice-bot-automation</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=3ba323070d49" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How n8n Helped Us Automate Monthly Employee Leave Updates]]></title>
            <link>https://medium.com/@ashwini.todewar/how-n8n-helped-us-automate-monthly-employee-leave-updates-b83c8f16a3ae?source=rss-cbcae1e345c4------2</link>
            <guid isPermaLink="false">https://medium.com/p/b83c8f16a3ae</guid>
            <category><![CDATA[playwrights]]></category>
            <category><![CDATA[n8n]]></category>
            <category><![CDATA[jenkins]]></category>
            <category><![CDATA[n8n-workflow-automation]]></category>
            <dc:creator><![CDATA[Ashwini Todewar]]></dc:creator>
            <pubDate>Wed, 05 Aug 2026 08:52:44 GMT</pubDate>
            <atom:updated>2026-08-05T08:52:44.003Z</atom:updated>
            <content:encoded><![CDATA[<p>Manual processes often start small and manageable, but as teams grow, they quickly become time-consuming and difficult to maintain. We faced a similar challenge with monthly employee leave updates across multiple project spreadsheets.</p><p>Every month, someone had to review project sheets, identify which ones needed updates, collect employee leave information, update records manually, and share the updated sheets with stakeholders. While none of these tasks were particularly complex, they consumed valuable time and introduced the risk of human error.</p><p>To solve this, we decided to automate the entire process using n8n.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*5SCcTTMzMwMPx_ksLMlfyQ.png" /></figure><h3>Why n8n?</h3><p>We needed a solution that could connect multiple systems, run on a schedule, and handle both success and failure scenarios without requiring constant monitoring.</p><p>n8n stood out because it offered everything we needed in a single platform:</p><ul><li>Scheduled workflow execution</li><li>Google Drive integration</li><li>Google Sheets integration</li><li>API and HTTP support</li><li>Built-in looping and data transformation</li><li>Automated notifications</li></ul><p>Instead of maintaining multiple scripts and manual processes, we could manage the complete workflow through one visual automation platform.</p><h3>The Workflow</h3><p>The automation starts with a monthly Schedule Trigger in n8n.</p><p>At the scheduled time, n8n scans a Google Drive folder containing project-specific employee sheets. Before processing any file, the workflow checks whether the sheet has already been modified. If updates were already made manually, the sheet is skipped to prevent accidental overwrites.</p><p>After identifying the sheets that require processing, n8n extracts the project names and prepares them for the next step.</p><p>The workflow then triggers a Jenkins job and passes the project list as a parameter. Jenkins acts as the execution layer, where a <a href="https://github.com/ashwinitodewar491/AutomatedInvoiceGeneration">Playwright automation framework</a> is responsible for collecting the latest employee leave information from the source application. Once the Playwright execution is complete, it generates an artifact containing employee leave details for all requested projects.</p><p>While the job is running, n8n continuously monitors the Jenkins build status and waits until execution is complete. This ensures that downstream steps only run when fresh data is available.</p><p>Once the build succeeds, n8n retrieves the generated artifact and extracts the employee leave data. Before applying any updates, it performs another validation check on the target sheets. This additional safeguard ensures that no manual changes made during workflow execution are overwritten.</p><p>The employee data is then processed and matched with the corresponding project sheets. Using loops and Google Sheets nodes, n8n updates employee leave records automatically, one employee at a time.</p><p>Finally, the workflow sends a summary email containing links to all updated sheets. If any step fails — whether it’s the Jenkins build, Playwright execution, or sheet update process — a failure notification is sent with relevant details, allowing the issue to be investigated quickly.</p><h3>The Impact</h3><p>The biggest benefit of this automation was not simply reducing a few manual steps. It completely removed a recurring monthly administrative task.</p><p>What previously required manual verification, spreadsheet updates, report collection, and email communication now runs automatically with minimal supervision.</p><p>The workflow improved accuracy, reduced the risk of human error, and ensured that employee leave records remained consistent across projects. The built-in validation checks also protected manually updated sheets from being overwritten by automation.</p><p>Most importantly, the process became reliable and scalable. Whether there were a handful of projects or dozens of them, the workflow handled the process consistently without additional effort.</p><h3>Final Thoughts</h3><p>This project reinforced an important lesson: automation is not always about replacing complex processes. Sometimes the biggest gains come from eliminating small, repetitive tasks that consume time month after month.</p><p>By using n8n as the central orchestration platform, we connected Google Drive, Google Sheets, Jenkins, Playwright, and email notifications into a single automated workflow. n8n handled the scheduling, validation, monitoring, data processing, and communication, while Jenkins and Playwright focused on generating the required employee leave data.</p><p>The result was a reliable, scalable solution that transformed a manual monthly process into a hands-off operation. What once required multiple manual steps now runs automatically every month, allowing the team to focus on more valuable work instead of spreadsheet maintenance.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b83c8f16a3ae" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Why I Built QaBoilerPlate: One Repository for Every Modern QA Engineer]]></title>
            <link>https://medium.com/@ashwini.todewar/why-i-built-qaboilerplate-one-repository-for-every-modern-qa-engineer-12eee195a06f?source=rss-cbcae1e345c4------2</link>
            <guid isPermaLink="false">https://medium.com/p/12eee195a06f</guid>
            <category><![CDATA[rules]]></category>
            <category><![CDATA[qa]]></category>
            <category><![CDATA[playwright-automation]]></category>
            <category><![CDATA[boilerplate]]></category>
            <category><![CDATA[ai-skills]]></category>
            <dc:creator><![CDATA[Ashwini Todewar]]></dc:creator>
            <pubDate>Wed, 29 Jul 2026 12:17:30 GMT</pubDate>
            <atom:updated>2026-08-14T09:34:33.727Z</atom:updated>
            <content:encoded><![CDATA[<p>Every automation project starts with the same routine — setting up folders, configuring dependencies, creating a framework structure, and writing common utilities before the first test is even automated.</p><p>After doing this repeatedly across different projects, I started asking myself:</p><p><strong>Why rebuild the same foundation every time?</strong></p><p>The more automation projects I worked on, the more I realized that a significant amount of QA engineering time is spent solving the same setup and architecture problems again and again.</p><p>That question led me to create <strong>QaBoilerPlate</strong> — a reusable, AI-ready QA automation starter designed to help QA engineers start automation projects faster while following consistent engineering standards.</p><h3>One Repository. Multiple Automation Stacks.</h3><p>QA teams work with different technologies depending on the project. Some applications still rely heavily on Selenium, while others are moving toward Playwright. JavaScript and TypeScript ecosystems have also made Cypress a popular choice.</p><p>Instead of maintaining completely separate starter repositories for each technology, QaBoilerPlate brings multiple automation stacks together under one structure.</p><p>It currently provides foundations for:</p><ul><li><strong>Selenium</strong> — UI automation with Java and TestNG</li><li><strong>Playwright</strong> — UI and API automation with Java and TestNG</li><li><strong>Cypress</strong> — UI and API automation using JavaScript/TypeScript</li></ul><p>The idea is simple:</p><p><strong>Choose your technology, start from an established foundation, and focus on writing meaningful tests instead of spending hours setting up the project.</strong></p><p>This can be especially useful when teams need to experiment with a new automation technology or migrate an existing suite from one framework to another.</p><h3>More Than Just Frameworks</h3><p>Most automation starter repositories focus primarily on code.</p><p>But a good automation framework is not only about folders, classes, and dependencies.</p><p>It is also about <strong>engineering practices</strong>.</p><p>QaBoilerPlate includes reusable QA Rules designed to encourage consistency across automation projects. These rules cover areas such as:</p><ul><li>Locator strategy</li><li>Framework architecture</li><li>Assertion best practices</li><li>Test data management</li><li>Coding standards</li><li>Reusable automation patterns</li><li>Maintainability and scalability</li></ul><p>Why does this matter?</p><p>Because two engineers can use the same automation framework and still produce completely different code.</p><p>One engineer might create stable locators and reusable methods, while another might introduce hardcoded values, duplicated code, unnecessary waits, and tightly coupled test cases.</p><p>Over time, these small differences become expensive to maintain.</p><p>Having common engineering rules provides a shared baseline for the team and makes automation easier to understand, review, and maintain.</p><h3>Built for the AI Era</h3><p>AI coding assistants are rapidly becoming part of everyday development and QA workflows.</p><p>Tools such as ChatGPT, Claude, Cursor, and GitHub Copilot can generate test cases, page objects, API clients, utilities, and even complete automation flows.</p><p>But there is a problem.</p><p><strong>AI can generate code quickly, but it does not automatically know your team’s engineering standards.</strong></p><p>Without proper guidance, AI-generated automation can introduce inconsistent naming conventions, duplicate utilities, poor locator strategies, unnecessary waits, or framework patterns that don’t match the existing project.</p><p>That is why QaBoilerPlate is designed to be <strong>AI-ready</strong>.</p><p>The repository includes AI Skills and rules that can guide AI assistants toward generating automation aligned with the framework’s architecture and QA standards.</p><p>Instead of repeatedly explaining:</p><blockquote><em>“Use this locator strategy.”</em></blockquote><blockquote><em>“Follow this folder structure.”</em></blockquote><blockquote><em>“Don’t duplicate this utility.”</em></blockquote><blockquote><em>“Use this assertion approach.”</em></blockquote><p>You can provide the AI tool with the project’s established rules and allow it to work within those boundaries.</p><p>The goal isn’t to replace the QA engineer.</p><p>The goal is to make the QA engineer <strong>more productive</strong>.</p><p>AI handles more of the repetitive work, while the engineer focuses on test strategy, risk analysis, quality, and solving the problems that actually require human judgment.</p><h3>Why I Built It</h3><p>My goal wasn’t to create another automation framework.</p><p>There are already excellent frameworks available for UI and API automation.</p><p>The real problem I wanted to solve was the <strong>repetitive foundation work that happens before automation begins</strong>.</p><p>Every new project usually requires decisions around:</p><ul><li>Project structure</li><li>Dependency management</li><li>Test execution</li><li>Reporting</li><li>Utilities</li><li>Test data</li><li>Coding standards</li><li>Framework patterns</li><li>CI/CD integration</li></ul><p>These decisions don’t necessarily need to be reinvented for every project.</p><p>QaBoilerPlate is my attempt to create a reusable starting point where those decisions can evolve into a common engineering foundation.</p><p>The goal is to help QA engineers:</p><ul><li>Start automation projects faster</li><li>Follow consistent engineering practices</li><li>Work across Selenium, Playwright, and Cypress</li><li>Build automation that scales</li><li>Reduce repetitive setup work</li><li>Collaborate more effectively with AI tools</li><li>Spend more time on quality and less time on boilerplate</li></ul><h3>What’s Next?</h3><p>QaBoilerPlate is not intended to be a finished product.</p><p>Automation engineering continues to evolve rapidly, especially with AI becoming part of the development and testing lifecycle.</p><p>The repository will continue evolving with new automation patterns, reusable utilities, QA engineering practices, AI Skills, reporting approaches, and framework improvements.</p><p>I also want it to become something that other QA engineers can contribute to rather than something that remains a personal collection of automation templates.</p><p>If you have ideas for improving the architecture, adding another technology, introducing better QA practices, or making AI-assisted automation more reliable, contributions and feedback are welcome.</p><h3>Explore QaBoilerPlate</h3><p>If you’re a QA engineer who frequently starts new automation projects, experiments with different automation technologies, or wants a reusable foundation for AI-assisted automation, you can explore the repository here:</p><p><strong>GitHub Repository:</strong><br><a href="https://github.com/ashwinitodewar491/QaBoilerPlate">https://github.com/ashwinitodewar491/QaBoilerPlate</a></p><p>Because great automation isn’t just about writing tests.</p><p><strong>It’s about building a foundation that helps every future project succeed.</strong></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=12eee195a06f" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Real Impact of AI on QA Engineering: Beyond Test Generation]]></title>
            <link>https://medium.com/@ashwini.todewar/the-real-impact-of-ai-on-qa-engineering-beyond-test-generation-7cabd41e630c?source=rss-cbcae1e345c4------2</link>
            <guid isPermaLink="false">https://medium.com/p/7cabd41e630c</guid>
            <category><![CDATA[ai-skills]]></category>
            <category><![CDATA[anthropic-claude]]></category>
            <category><![CDATA[testingrule]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Ashwini Todewar]]></dc:creator>
            <pubDate>Mon, 08 Jun 2026 07:06:36 GMT</pubDate>
            <atom:updated>2026-06-08T07:06:36.405Z</atom:updated>
            <content:encoded><![CDATA[<p>When AI first entered the testing space, most discussions focused on one question:</p><p><strong>Can AI generate automation tests?<br></strong>The answer is <strong>yes</strong>.</p><p>Tools like Claude, Cursor, and GitHub Copilot can generate Cypress tests within seconds. However, after experimenting with these tools, I realized that generating a test is rarely the hard part.</p><p>The real challenge is generating tests that remain <strong>stable, maintainable, and aligned with framework standards</strong> months later.</p><p>This is where I believe the real impact of AI on QA Engineering begins — not as a test generator, but as an <strong>automation governance partner</strong>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*4YAYMsY4aY8rIXF6zxHOeA.png" /></figure><h3>The Problem with AI-Generated Tests</h3><p>Let’s take a simple example.</p><h3>Prompt</h3><p>Create a Cypress test to save a presentation.</p><p>A typical AI-generated response might look like this:<br>cy.get(‘body &gt; div &gt; div &gt; button’).click()<br>or<br>cy.get(‘.css-123abc’).click()</p><p>The test works.<br>The locator strategy doesn’t.</p><p>These selectors are fragile and highly dependent on the application’s structure. Any UI change can break the test.</p><p>Without guidance, AI simply generates code based on what it sees in the DOM. It has no understanding of your team’s automation standards.</p><h3>The Missing Piece: Rules</h3><p>Instead of treating AI as a code generator, we can teach it how our framework works.</p><p>This is the idea behind the repository I am building:<br>cypress-ai-governance</p><p>The goal is simple:<br>Teach Claude, Cursor, and Copilot to generate maintainable Cypress automation by following predefined framework rules.</p><h3>Teaching AI a Locator Strategy</h3><p>One of the first areas where AI can help is locator quality.<br>For example, we can define rules such as:</p><h3>Preferred Locator Order</h3><ol><li>data-testid</li><li>findByRole</li><li>findByLabelText</li><li>findByText</li></ol><h3>Avoid</h3><ul><li>XPath</li><li>nth-child</li><li>Dynamic CSS classes</li><li>body &gt; div chains</li><li>parent().parent().parent()</li></ul><p>Now when AI encounters:<br>cy.get(‘body &gt; div &gt; div &gt; button’)</p><p>it should not simply generate the code.</p><p>Instead, it should respond:<br>⚠ Brittle Locator Detected</p><p>Reason:<br>Depends on DOM structure.</p><p>Recommendation:<br>Add data-testid=”save-button”</p><p>Suggested Locator:<br>cy.get(‘[data-testid=”save-button”]’)<br>At this point, AI is acting more like a reviewer than a generator.</p><h3>Moving Beyond Self-Healing</h3><p>The industry talks extensively about self-healing locators.<br>While self-healing is useful, I believe prevention is more valuable.<br>Rather than fixing broken locators after a test fails, AI can help identify fragile selectors during test creation.</p><p>Examples of patterns that should be flagged automatically:<br>→ cy.get(‘.css-123abc’)</p><p>→ cy.get(‘body &gt; div &gt; div &gt; button’)</p><p>→ cy.get(‘.card’)<br>.parent()<br>.parent()<br>.find(‘button’)</p><p>These selectors may work today, but they introduce long-term maintenance costs.<br>AI can detect these patterns and recommend more stable alternatives before the code ever reaches a pull request.</p><h3>Using JSDoc Warnings</h3><p>There are situations where a stable locator simply does not exist.<br>Instead of hiding the risk, AI can document it.<br>Example:<br>/**<br>* WARNING:<br>* Locator depends on CSS class.<br>* Consider adding data-testid.<br>*/<br>cy.get(‘.btn-primary’).click()</p><p>This provides visibility and helps teams track technical debt inside their automation suite.</p><h3>Framework Governance</h3><p>Locators are only one part of maintainability.<br>AI can also enforce framework architecture.<br>For example, in our framework we may define:</p><h3>Framework Rules</h3><ul><li>Use Page Objects</li><li>Avoid raw selectors in spec files</li><li>Reuse custom commands</li><li>Reuse helper methods</li></ul><p>Instead of generating:<br>cy.get(‘[data-testid=”save-button”]’).click()</p><p>AI should generate:<br>presentationPage.clickSaveButton()<br>This ensures consistency across the automation framework.</p><h3>Test Data Governance</h3><p>Another common issue in automation projects is hardcoded test data.<br>For example:<br>cy.get(‘#email’)<br>.type(‘test@gmail.com’)</p><p>Rules can instruct AI to:</p><ul><li>Use Faker</li><li>Use data factories</li><li>Use API setup</li><li>Avoid hardcoded values</li></ul><p>Result:<br>const email = faker.internet.email()<br>contactPage.enterEmail(email)<br>This makes generated tests more reliable and reusable.</p><h3>Assertion Governance</h3><p>AI can also enforce better assertion practices.</p><p>Instead of:<br>cy.wait(5000)</p><p>AI should recommend:<br>cy.findByText(‘Saved Successfully’)<br>.should(‘be.visible’)</p><p>This improves stability while reducing unnecessary execution time.</p><p>Find following Repo Link for how we can add rules to repo and train AI: <a href="https://github.com/ashwini-todewar/claude-cypress-rules">https://github.com/ashwini-todewar/claude-cypress-rules</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=7cabd41e630c" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Why Shift Left Testing: What Actually Works in Agile Teams]]></title>
            <link>https://medium.com/@ashwini.todewar/why-shift-left-testing-what-actually-works-in-agile-teams-533997181bf4?source=rss-cbcae1e345c4------2</link>
            <guid isPermaLink="false">https://medium.com/p/533997181bf4</guid>
            <category><![CDATA[testing]]></category>
            <category><![CDATA[agile]]></category>
            <category><![CDATA[qa]]></category>
            <dc:creator><![CDATA[Ashwini Todewar]]></dc:creator>
            <pubDate>Mon, 01 Jun 2026 10:48:10 GMT</pubDate>
            <atom:updated>2026-06-01T10:48:10.176Z</atom:updated>
            <content:encoded><![CDATA[<p>Shift Left Testing is one of those terms we hear frequently in Agile teams. For a long time, I thought it simply meant starting automation earlier in the sprint.</p><p>Over the years, I’ve realized that successful Shift Left Testing is much more than automation. It’s about building quality into every stage of development instead of waiting for QA to find defects at the end.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*V0StUGgx92B8Ez3TK2HXvw.png" /></figure><h3>What Is Shift Left Testing?</h3><p>In a traditional approach, testing happens after development is complete:</p><p>Requirements → Development → QA Testing → Release</p><p>The problem is that defects discovered late are more expensive to fix and often delay releases.</p><p>Shift Left Testing encourages teams to focus on quality much earlier:</p><p>Requirements → Review → Development → PR Validation→ Manual testing→ Feature stability → Start Automation → Release</p><p>The goal is simple: catch issues before they become defects.</p><h3>1. Start During Requirement Discussions</h3><p>One of the easiest ways to prevent defects is by asking questions before development begins.</p><p>For example, if a requirement says:</p><p><em>“Users can share presentations.”</em></p><p>A QA engineer may ask:</p><ul><li>Who can share them?</li><li>Can they be shared externally?</li><li>Are there permission restrictions?</li><li>Should shared links expire?</li></ul><p>These discussions often uncover missing requirements before any code is written.</p><h3>2. Design Test Scenarios Early</h3><p>Instead of waiting for development to finish, QA can prepare test scenarios while the feature is being built.</p><p>For a CSV upload feature, possible scenarios might include:</p><ul><li>Valid file upload</li><li>Empty file</li><li>Invalid format</li><li>Duplicate records</li></ul><p>Having these scenarios ready early helps teams validate features faster once development is complete.</p><h3>3. Participate in Pull Request Reviews</h3><p>QA involvement in Pull Request reviews can significantly improve quality.</p><p>This doesn’t mean reviewing every line of code. Instead, focus on questions such as:</p><ul><li>Does this change impact existing automation?</li><li>Are new edge cases introduced?</li><li>Is additional test coverage needed?</li></ul><p>A simple discussion during a PR review can prevent defects from reaching QA environments.</p><h3>4. Use Automation as a Quality Gate</h3><p>Automation is most effective when integrated into CI/CD pipelines.</p><p>Common checks include:</p><ul><li>Unit tests</li><li>API tests</li><li>Smoke tests</li><li>Static code analysis</li></ul><p>The earlier developers receive feedback, the faster issues can be resolved.</p><h3>5. Make Quality a Shared Responsibility</h3><p>One common misconception is that Shift Left Testing is solely a QA initiative.</p><p>In reality, quality belongs to the entire team:</p><ul><li>Product Managers provide clear requirements.</li><li>Developers write reliable code and unit tests.</li><li>QA engineers identify risks and automate validations.</li><li>DevOps teams ensure quality checks run in pipelines.</li></ul><p>When everyone contributes, fewer defects reach production.</p><h3>6. The Cost of Finding Defects Late</h3><p>One important lesson I’ve learned throughout my QA journey is that the cost of fixing a defect increases significantly as it moves through the software development lifecycle.</p><p>A missing requirement identified during a grooming or refinement session may take only a few minutes to clarify.</p><p>The same issue discovered during QA testing could require code changes, additional validation, regression testing, and another deployment cycle.</p><p>If that defect reaches production, the impact becomes much greater. It may affect customers, generate support tickets, require urgent fixes, and potentially impact business reputation.</p><p>This is one of the key reasons why Shift Left Testing is so valuable. By identifying risks and gaps earlier in the process, teams can reduce rework, improve delivery speed, and minimize production issues.</p><p>In my experience, the most effective teams are not necessarily the ones that find the most defects — they are the ones that prevent defects from being introduced in the first place.</p><p>That’s why Shift Left Testing is more than a testing strategy; it’s a business advantage.</p><h3>Final Thoughts</h3><p>Shift Left Testing is not about moving testing activities earlier — it is about building quality into the development process from the beginning.</p><p>The practices that have worked best for me are:</p><ul><li>Requirement reviews</li><li>Early test design</li><li>PR validation</li><li>CI/CD automation</li><li>Fast feedback loops</li></ul><p>The most successful teams are not the ones that find the most bugs. They are the teams that prevent those bugs from being created in the first place.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=533997181bf4" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>