The age of agentic testing

Unlocking Quality: How Agentic Testing Reshapes Software Development at Web Summit Lisbon 2025

Mohammad Asad Khan, Co-founder and CEO of LambdaTest, addressed Web Summit Lisbon 2025, highlighting a critical shift in software engineering. With Large Language Models (LLMs) generating code rapidly, the vital aspect of “white testing” is often overlooked. He emphasized the dangers of neglecting robust testing in this new AI-driven development era, advocating for a proactive approach to quality.

LambdaTest, a six-year-old company, has evolved from a test execution platform to a comprehensive AI quality engineering platform, leveraging agentic systems. Backed by major investors, LambdaTest is expanding. The landscape now features diverse AI agents, from task-specific tools to autonomous, multi-modal systems. Enterprises are embracing AI, but testing these complex, non-deterministic systems presents a significant challenge.

Traditionally, software development focused 80% on coding and 20% on testing. However, with AI accelerating code shipment, this paradigm is flipping, demanding an 80% testing to 20% code ratio. Testing, once reactive, must become proactive and continuous to match hourly release cycles. Manual methods are insufficient; autonomous agents are now essential for ensuring quality at scale.

LambdaTest’s platform integrates AI across the testing workflow. LLMs generate 70-80% of test plans and cases from various inputs. Human testers then refine the reasoning, adding critical insights. KAI, LambdaTest’s agent, enables natural language automation, allowing thousands of tests to be created overnight, a task that would take human engineers months.

For execution, LambdaTest provides a high-speed agentic infrastructure. It automatically splits and parallelizes test suites across isolated nodes, preventing bottlenecks. The system features auto-healing capabilities, identifying and fixing issues, and providing root cause analysis (RCA) for quick debugging within CI/CD pipelines. This ensures continuous, efficient testing and reduces time spent triaging issues.

The role of testers is evolving from merely verifying “does it work?” to assessing “does it reason well?”. This shift emphasizes context engineering and continuous training of AI systems. While AI agents won’t replace human testers soon, their role becomes more strategic, focusing on ensuring the trustworthiness and ethical reasoning of mission-critical software, where human oversight remains crucial.

LambdaTest pioneers agent-to-agent testing, a “battle of bots” where one agent ships code while another, a synthetic agent, actively seeks vulnerabilities. This mechanism supports testing across chat, voice, and phone-to-phone interactions. The system collects multi-modal requirements and defines user personas to evaluate agents comprehensively, assessing conversation flow, user satisfaction, empathy, and professionalism.

Evaluation metrics include assessing toxicity, hallucination, and biased behavior for chat agents, and parameters like intent recognition, latency, and words per minute for voice agents. These metrics combine with persona-based scoring to provide a holistic trust score. A real-world example showed an agent revealing an AWS key, underscoring the critical need for robust security guard rails.