AI Is Testing AI-Generated Code: Why You Should Trust It
Note: This article was originally published on July 28, 2025, and has been updated with information from the webinar.
That AI has become the engine behind modern software development is undeniable: More than 30% of Google’s code is written by AI; at Microsoft, the number is “as much as 30%.”
While generative AI (GenAI) can write more code more quickly than human developers, making people feel more productive, it introduces significant problems: Because AI-written code is unpredictable, often inaccurate, unsafe and unreliable, testing and debugging require significant effort.
As Deba Chatterjee, head of products for AI Test Automation at Harness AI, explained in a recent webinar, “you may encounter a production issue that you haven’t encountered before.” One reason is that the way you test human-written code is very different from how you can test AI-generated code, Chatterjee said, which potentially creates reliability and security issues.
“When you bring a developer to troubleshoot a problem that was created off of AI-generated code, the developer has to reverse-engineer it, and then figure out where the issues are,” he said. This tracks with a key finding in Harness’ State of Software Delivery 2025 report: two-thirds of developers say they spend excessive time debugging and resolving security vulnerabilities in AI-generated code.
As AI becomes the engine behind modern software delivery, ensuring the reliability and safety of AI-driven workflows is mission-critical.
Why AI Might Be the Right Answer
As organizations try to solve these problems, many are exploring ways to use GenAI to test their code, both AI and non-AI workflows. It seems counterintuitive: Why would you use unpredictable technology to test unpredictable technology?
Fortunately, this is not only possible, but companies like Harness are already successfully putting it into practice. And by watching AI Testing AI: How to Build Unbreakable Validation for Unpredictable Systems on demand, you’ll have the opportunity to see exactly how Harness is using AI-powered test generation, self-healing test suites, and intelligent test selection to validate AI workflows.
During this free webinar, Chatterjee and Rohan Gupta of Harness joined The New Stack Host Chris Pirillo to pull back the curtain on how Harness uses advanced agentic AI-powered testing to rigorously validate its own AI workflows.
They not only demonstrated the power of AI in testing AI; they explored how to use agentic AI-powered testing to test non-AI workflows.
Watch This Free Webinar Today!
What You’ll Learn
By watching this webinar on demand, you’ll receive best practices, real-world examples and actionable tips including:
- Leveraging AI-powered test generation, self-healing test suites and intelligent test selection to achieve rapid, resilient and scalable validation of AI workflows.
- Using advanced techniques to comprehend, challenge and validate the intricate, dynamic behaviors of AI workflows, thereby ensuring trust and transparency in automated decision-making.
- Seeing how using AI to test AI has slashed test cycle times by up to 80%, reduced test maintenance by 70% and enabled continuous quality at the speed of innovation.
Chatterjee and Gupta also demonstrated how Harness’s AI Test Automation validates AI and non-AI workflows, showcasing intent-based testing, smart selectors and automated adaptation to application changes. Watch now and learn what you need to know about AI testing AI!
Watch this on-demand webinar now!
