Systems | Development | Analytics | API | Testing

Generative AI for QA is the use of large language models to accelerate the creation and analysis of testing artifacts — drafting test cases, summarizing requirements, and generating synthetic test data. AI agents extend that capability into multi-step autonomous workflows that plan, delegate, and execute testing tasks across an entire delivery pipeline. For SDETs, the shift is not about learning to prompt more cleverly.

The question isn't whether to use AI in QA. It's knowing exactly where to keep a human in control. The core risk: Over 75% of multi-agent failures are silent semantic errors that pass automated checks but violate business logic — detectable only by human inspection (Cemri, Pan et al., NeurIPS 2025). The division of labor: AI owns repetitive generation and execution; humans own risk analysis, requirement interpretation, exploratory investigation, and final sign-off. The operational discipline.

Every enterprise contract I’ve signed in the last two years has the same clause. “Vendor will not use Customer Data to train machine learning models.” Sometimes it’s a paragraph. Sometimes it’s a whole section. The language varies but the intent is identical: don’t feed our production data into your AI. I get it. I sign the same clause as a vendor. But here’s what’s been bothering me: that clause is a promise, not an architecture.

The most dangerous multilingual AI failure is a fluent output that feels native enough to trust, but is wrong enough to mislead.

As AI-aided software development lifecycles pick up speed, performance teams are left with the familiar challenge of too much work, too few specialists, and results that take too long to analyze. Over the past year, Tricentis NeoLoad has shipped capabilities designed to address each of these problems directly. What started with Augmented Analysis accelerating root cause identification grew into a fully connected Model Context Protocol (MCP) architecture.

At Ably we've been able to make a transformational change in how we build SDKs as a result of what's now possible with LLMs. In this post, I explain how.

Writing test cases sounds simple until you're actually doing it. You read a requirement, try to figure out what "done" even means for it, write out steps, realize you missed three scenarios, go back, revise, and by the time you feel okay about it, the sprint has moved on without you. This happens to every team working at any real scale. It is just what manual test case creation costs you, and it is the problem autonomous test generation was supposed to solve at a category level.