Regression testing often grows silently until the safest answer becomes “run everything again.” That feels responsible, but it can burn time, infrastructure, and attention without improving confidence. In this practical talk, Dr. Anubha Jain shows how teams can move from volume-driven regression to decision-driven testing. Based on a Salesforce AI-enabled system with more than 1,200 regression tests running across four environments, she explains how the team questioned the habit of full-suite execution several times a week. The solution combined change-impact analysis, risk-based selection, historical failure signals, and selective AI prioritization, not blind automation on top of a broken process. The result was a smaller, smarter execution set of around 700 high-impact tests per cycle, with more than 40 percent fewer executions and preserved release confidence. The session also connects this work to sustainability: fewer pointless runs mean less compute waste, lower cost, shorter feedback loops, and better focus for testers and developers. Attendees will learn what signals to use, how to challenge regression habits, and how to talk about efficiency without sounding like they want to cut corners. The key message is simple - sustainable testing starts when testers stop acting like executors and start acting like decision-makers.

With the rise of AI-powered, agentic systems, it is tempting to believe that established automation frameworks are becoming yesterday’s news. Why invest in architecture, page objects, stable locators, and explicit assertions when an intelligent agent can supposedly “understand” the user interface and adapt on its own? Nikolay Avramov brings this question back to earth through his experience building a custom agentic testing framework that tried to execute tests based on intent rather than predefined scripts. The promise was attractive: less maintenance, self-healing behavior, and a shift from code to natural language. The reality was more complicated. Without the discipline of a strong testing framework, agentic systems can amplify flakiness instead of removing it. Ambiguous intent leads to inconsistent decisions, weak assertions create false confidence, and non-deterministic behavior becomes difficult to trust. Through real examples, this talk explores where AI-based testing adds value, where it breaks down, and why tools like Atata, Bellatrix, or Appium-based solutions still matter. It is especially useful for teams trying to decide whether AI should replace parts of their automation stack or sit on top of it. The answer is not “frameworks are dead.” The answer is that AI needs structure, boundaries, and clear expectations. AI does not remove testing discipline - it raises the bar.

AI is changing software delivery fast enough to make job titles feel a little outdated. Tools can now generate test ideas, explain failures, write code, suggest fixes, and challenge requirements before a ticket even reaches implementation. That does not mean QA disappears. It means the QA role moves closer to product thinking, risk analysis, technical understanding, and decision-making. In this talk, Dimitar Topuzov, known to many in the community as Uncle Mitko, looks at the shift from QA engineer to AI-powered product engineer in a practical and honest way. He will explore how the borders between QA, developer, product manager, and UX specialist are becoming softer, and why the tester mindset is still valuable exactly because of that change. The session will look at what skills become more important when tools can produce output quickly: asking better questions, validating assumptions, understanding users, spotting weak requirements, and deciding where human review is still needed. Expect observations from real engineering work, a few uncomfortable questions, and a grounded view of what AI can and cannot do for quality. The session is for people who want to stay relevant without panicking every time a new tool promises to replace half the team.

Whenever AI enters a QA conversation, the first question is usually: will it take our jobs? Alla Penalba approaches that question through a real transformation story, not through predictions from a crystal ball. She will share how a QA organization embedded agentic AI across the full software development lifecycle - from discovery and idea validation, through requirements and specifications, to execution, reporting, and release support. This was not a side experiment or a fancy demo. It was a change in how QA work gets done when delivery speed grows and traditional manual testing starts becoming the bottleneck. The talk will show what worked, what failed, and where human judgment remained essential. It will also explore how the QA engineer’s role changes from executor to validator, reviewer, and quality strategist. For QA managers and engineers, this is a practical map for adopting AI without losing control of quality. Alla will also discuss starting points for teams at different maturity levels, because not every organization can jump directly into a fully AI-assisted workflow. The session keeps the focus on real outcomes, not buzzwords: better decisions, faster feedback, and stronger ownership. The future may belong to the QA Cyborg, but someone still needs to teach the machine what good looks like.

Testing applications powered by Large Language Models (LLMs) is not just regular testing with a smarter text box. These systems are probabilistic, sensitive to prompts, and sometimes confidently wrong in ways that traditional expected-result checks cannot handle. Fabio Leite will walk through the practical challenges of testing LLM-based applications, including hallucinations, bias, inconsistency, safety, relevance, and prompt sensitivity. The session will also clarify the difference between standard LLM applications and Retrieval-Augmented Generation (RAG) systems, where answers depend not only on the model, but also on retrieved context and grounding quality. Fabio will introduce modern evaluation approaches such as LLM-as-a-Judge, human-in-the-loop review, and basic automation patterns that help teams build confidence without pretending the system is deterministic. The talk avoids vendor lock-in and focuses on mental models QA teams can apply across tools and platforms. Attendees will learn how to think about accuracy, consistency, safety, and relevance as test dimensions, and how to create evaluation workflows that are useful even when output varies. No deep AI background is required, but the talk will be valuable for QA engineers, developers, tech leads, and product people working with GenAI features. QA has a critical role in making AI products trustworthy, especially when “expected output” is no longer a single fixed string.

Testing for the Olympic Winter Games is not exactly a calm little side project. It means heavy data flows, strict deadlines, many stakeholders, changing requirements, and the tiny detail that the whole world is watching. Marin Ivanov turns the most stressful nine months of his career into a practical and humorous QA case study about B2B systems, sports data, and staying sane when every second counts. He will share the DOs and DON’Ts of working on a high-pressure project connected to one of the biggest events in the sports calendar. The talk is especially useful for QA engineers, team leads, and project managers who work close to developers and clients, where the problem is rarely just “find the bug.” It is also about understanding data, managing expectations, communicating clearly, and knowing which risks deserve attention before the deadline hits like a curling stone. Marin will cover what needs to happen before such a project starts, what to watch during the critical delivery period, and what lessons should be captured after the dust settles. Expect real lessons, not superhero stories, plus a practical checklist for staying sharp when quality, data, clients, and deadlines all collide.

Modern systems are too complex for test suites to know everything in advance. We write scripts for the known risks, but many production failures live in the unknown-unknowns: unexpected user journeys, service interactions, latency patterns, data conditions, and AI behaviors that never appeared in the lab. David Burns introduces Observability-Driven Testing as a way to connect testing with the high-fidelity data already flowing through modern systems. Instead of treating test environments and production as completely separate worlds, this approach uses telemetry, traces, and real behavior as a smarter test oracle. The session will explore how OpenTelemetry (OTel), trace-based testing, and real-user journeys can help teams validate system correctness beyond hard-coded assertions and mock data. This becomes even more important as software moves toward non-deterministic AI agents, micro-frontends, and distributed architectures where the “expected result” is often a pattern, not a static value. David will show how traces can inspire automated integration checks, how production signals can sharpen test design, and how teams can use observability without turning it into another expensive dashboard nobody reads. Attendees will leave with a practical blueprint for moving from “Is the button green?” to “Is the system behaving the way users and production signals tell us it should?”

After a serious incident, companies rarely ask why they did not invest earlier in training, design reviews, pairing, risk analysis, or better testing strategy. They usually ask why QA did not catch it. Then money suddenly appears for hotfixes, mitigation, support, reputation repair, and all the expensive quality work that happens too late. Vitaly Sharovatov will show how the Cost of Quality model helps QA people reframe this painful conversation. Instead of trying to prove the invisible return on investment of prevention, teams can compare the cost of three choices for the same risk: keep absorbing failures, catch them earlier through appraisal and testing, or prevent them before they happen. The talk connects QA ethics with economics in a practical way, using risk-based thinking, ISO/IEC 25010 quality characteristics, incident data, and small Cost of Quality baselines. Vitaly will show how to start with one painful failure class, describe the risk clearly, define what “unacceptable” means, and propose a balanced portfolio of prevention, appraisal, and guardrails. It is especially useful for experienced testers, QA leads, and managers who need better arguments than “trust us, this is important.” The goal is to make quality investment visible, discussable, and easier to approve before the fire starts.

A product can pass functional tests and still fail the user. Buttons can work, APIs can respond, and test cases can be green, while the actual experience remains confusing, inaccessible, slow, or simply unpleasant. Pekka Pönkänen brings a product-quality perspective to a question many teams avoid: who tests user experience before the customer does? Drawing from years in product companies, native mobile development, and quality leadership at Wolt, this talk looks at quality as something broader than defect detection. It includes empathy, communication, accessibility, cybersecurity, and the courage to ask uncomfortable questions before users do it for you in reviews, support tickets, or social media. The session will explore how QA people can become stronger partners to product, design, and engineering teams, and how they can test not only what the system does, but how it feels to real people. Pekka will also connect this to feedback loops, release habits, and the small signals that reveal user pain before it becomes a public problem. It is a talk for testers, leads, and managers who believe that quality is not a department or a checklist - it is the complete user experience, including all the small paper cuts that never appear in a crash report.