In a recent LinkedIn post, Teresa Torres highlights an innovative approach to testing AI agents, as explained by Santi Marchiori of AITropos. Torres shares how Marchiori’s team has developed a sophisticated method for ensuring the quality and reliability of AI systems by using AI to test AI.
The core of this advanced testing strategy involves training a dedicated AI agent to act as a customer. This AI customer then engages in numerous interactions with the agent being tested. As Teresa Torres relays Marchiori’s explanation:
“We actually trained an agent that acts as a customer to test the agent. We run literally thousands of those during a night.”
This automated, high-volume testing is crucial for identifying potential issues that might arise in real-world scenarios. According to Teresa Torres, the process doesn’t stop after the initial interaction.
Automated Verification and Error Analysis
Torres explains that following the completion of a conversation between the AI customer and the agent under test, a secondary agent steps in to verify the accuracy of the order. This verification layer is critical for ensuring that the AI agent not only completes interactions but does so correctly and efficiently.
Furthermore, a third AI agent is employed to analyze errors across all the testing runs. This multi-layered AI oversight allows for a comprehensive understanding of performance and identifies areas for improvement. Teresa Torres points out the remarkable outcome of this methodology:
“What started with a huge error rate is now production-ready—built by AI testing AI.”
This statement underscores the transformative power of using AI for quality assurance, particularly in complex environments where traditional testing methods might fall short.
The Evolution of AI-Driven Quality Assurance
The insights shared by Teresa Torres through Marchiori’s work demonstrate a significant advancement in how AI systems are validated. By leveraging AI agents to simulate user behavior and meticulously check outcomes, the team at AITropos has managed to overcome initial high error rates and achieve a production-ready state for their AI.
According to Teresa Torres, this approach represents a paradigm shift in quality assurance. Instead of relying solely on human testers, which can be time-consuming and may not always capture the full spectrum of potential issues, an AI-driven system offers scalability, consistency, and speed. As Torres highlights, the entire process is an example of ‘AI testing AI,’ a concept that is becoming increasingly vital as AI adoption grows across industries.
The dedication to rigorous, automated testing, as detailed in the post covered by Torres, is essential for building trust and ensuring the reliability of AI applications. The journey from a high error rate to a production-ready system, facilitated by AI itself, is a testament to the potential of intelligent automation in refining AI performance.
📝 About This Content
This article is based on insights shared by Teresa Torres on LinkedIn.
📅 Originally posted on May 1, 2026 | View original post on LinkedIn →