Your build is green, coverage is 96%, every test passes—and the bug is already in production. The Tautology Trap occurs when AI generates both code and tests from the same requirement, allowing the same misunderstanding to be reproduced rather than detected. In this session, we’ll examine a real-world example where AI-generated code and tests achieve 100% coverage while shipping a costly defect, and explore why coverage can no longer be treated as a reliable quality signal.

The way out is metamorphic testing: testing relationships that must hold when inputs change, even when the correct answer is unknown. You’ll learn three practical metamorphic relations—Invariance, Direction, and Round Trip—and how to apply them to both traditional software and AI features. You’ll also leave with a simple principle for AI-assisted testing: humans own the invariants; AI owns the volume.

Software testing has evolved through client-server, web, mobile, cloud, automation, and DevOps. AI, however, raises a more fundamental question: what are we testing when software no longer has a single, predictable behavior? Drawing on three decades of experience building testing products, practices, and teams, Ashwin Palaparthi examines how concepts such as requirements, expected results, regression, coverage, and test oracles need to evolve—without losing the engineering discipline at the heart of testing.

Through practical examples, emerging patterns, and lessons from real-world AI applications and agents, this session explores how testers can reason about acceptable behavior, meaningful change, drift, uncertainty, and evidence of quality. Rather than introducing another evaluation framework or list of metrics, it challenges attendees to rethink what serious testing means when behavior itself becomes the moving target.

For decades, software followed a familiar model: people defined requirements, developers wrote code, testers validated it, and users consumed the result. AI is changing that model. Software is increasingly negotiated—between human intent and machine interpretation, speed and correctness, automation and accountability. Developers negotiate with AI-generated code, testers negotiate confidence, product teams negotiate business intent, and engineering leaders negotiate risk before software reaches production.

As AI takes over more execution, the value of software professionals shifts from producing outputs to shaping outcomes. This session explores how to challenge AI-generated assumptions, define effective constraints, ask better questions, and make sound engineering decisions. Through real-world examples and practical insights, discover why technical judgment, business context, and the ability to negotiate with intelligent systems may define the next decade of software engineering.

MCP (Model Context Protocol) servers introduce a new kind of client: an LLM that can interpret tool descriptions, generate unpredictable inputs, and invoke tools in unexpected sequences. Since we cannot reliably predict or control how an LLM will interact with an API, testing the client is not a practical CI strategy. The focus therefore shifts to what we can control—the MCP server. A well-designed server must be thoroughly tested to increase the likelihood of correct behavior from its non-deterministic clients.

This session presents a practical testing model for MCP servers, grounded in familiar testing principles but adapted to LLM-driven interactions. You’ll learn how to treat tool descriptions as functional contracts, test deliberate and out-of-order tool-call sequences, and validate error responses for effective LLM recovery. These techniques provide testers and developers with an immediately applicable approach to building resilient MCP servers as the ecosystem evolves.

Adopting AI tools for testing is easy. Proving that they actually improve quality, productivity, release confidence, and engineering efficiency is much harder. This session explores how teams can move beyond measuring AI-generated output and instead evaluate the outcomes that matter. We’ll simplify the growing AI testing landscape—including commands, skills, plugins, MCP servers, and agents—and demonstrate a real-world pipeline connecting Asana, Jira, Figma, TestRail, Playwright, and pull request reviews across the testing lifecycle.

From junior SDETs to senior engineers and engineering leaders, attendees will learn what to measure, how to measure it, and how to demonstrate meaningful impact. The session provides a practical approach to turning AI adoption into measurable improvements in quality, engineering effectiveness, and business value.

AI can now generate, execute, and maintain automated tests, promising faster delivery, broader coverage, and less maintenance. But what happens when AI becomes part of the testing system itself? Teams are discovering new challenges, including unpredictable test behavior, lost context across workflows, false confidence from passing tests, and difficulty understanding or debugging AI-driven decisions. These issues are no longer theoretical—they are emerging as teams scale AI-augmented automation.

This session explores what breaks when AI enters the testing lifecycle and, more importantly, what works. Through practical patterns and real-world lessons, attendees will learn where AI adds value, where human judgment remains essential, and how to build testing systems that are more reliable, observable, and resilient. The session also examines how QA roles and practices must evolve as AI becomes a core part of modern engineering.

AI is rapidly transforming software development and testing, but are organizations using it to improve Quality Engineering—or simply making existing QA processes faster, including the ineffective ones? AI can accelerate test creation, automation, execution, and analysis, but without strong leadership and governance, it can also amplify test noise, technical debt, security risks, and false confidence. This session takes a practical look at introducing AI across the Software Testing Life Cycle (STLC) while challenging practices such as “automate everything,” excessive test generation, and measuring AI adoption through output volume.

Attendees will learn practical approaches to identifying meaningful AI use cases, establishing governance and accountability, protecting enterprise data, evaluating AI-generated outputs, and measuring impact through engineering outcomes. The session also explores how QA roles are evolving and how leaders can reskill teams for AI-augmented Quality Engineering, intelligent automation, risk-based testing, and agentic workflows. The core message: don’t use AI to make a bad QA process faster; use it to make a good QA process smarter.