Sagar Joshi

Sagar Joshi is a software testing professional, trainer, and entrepreneur with over 16 years of experience in the automotive and enterprise software domains. He is the founder of SAJO Academy Private Limited, where he specialises in delivering practical, industry-aligned ISTQB® certification training for individuals and corporate teams.

Having worked with organisations such as Bosch, TCS, and KPIT, he brings strong expertise in test management, quality assurance, and agile practices. He holds multiple ISTQB certifications, along with PMP and Scrum credentials.

Sagar Joshi is also the author of Software Testing Essentials – An ISTQB® Foundation Guide and is driven by a clear mission: to help testers build strong fundamentals and grow sustainable careers through realistic, hands-on learning.

Traditional test strategies are built on assumptions that have served software testing well for decades: system behaviour is deterministic, expected results can be clearly defined, test data is neutral, and tools assist execution rather than judgement. AI-based systems quietly violate all of these assumptions. As a result, applying traditional test strategies to AI systems often creates a false sense of confidence rather than meaningful risk coverage.

This talk explains why testing AI systems breaks traditional test strategies — notbecause AI is more complex, but because it changes the nature of behaviour, decision-making, and knowledge within systems. AI models do not execute predefined logic; they infer behaviour from data. Their outputs are probabilistic, context-dependent, and often plausible even when incorrect. This leads to challenges such as the test oracle problem, hidden bias in test data, non-deterministic outcomes, model drift, and unclear accountability for decisions.

Using practical testing examples, including AI-generated test data and AI-assisted test design, the session shows how testers can unintentionally start treating AI output as “knowledge” rather than as hypotheses that require validation. When this happens, strategies based on expected results, coverage metrics, and regression stability begin to fail in subtle but critical ways.

Rather than promoting new tools or automation techniques, this session reframes AI testing as a strategy problem. It explores how testers must rethink risk, redefine acceptance criteria, and establish clear trust boundaries for AI outputs. The focus shifts from validating correctness to questioning assumptions and deciding when AI output is safe to use.

Attendees will leave with a clear understanding of where traditional test strategies break down for AI systems, why this matters, and how testers can adapt their thinking to test AI systems responsibly without being misled by convincing but unreliable results.