Patronus AI Unveils First Scalable Supervision Solution for Agentic Systems

As systems become increasingly autonomous, “Percival” provides AI oversight to automatically detect errors and optimize performance

SAN FRANCISCO, May 14, 2025 /PRNewswire/ — Patronus AI today unveiled Percival, the industry’s first self-serve AI solution that automatically identifies and suggests optimizations for agentic system failures. The tool addresses the growing challenge of maintaining reliable AI workflows as organizations scale their increasingly autonomous agent-based systems and applications.

AI systems have evolved from simple automation to autonomous agents that independently plan and execute complex tasks with minimal supervision. While this advancement has provided industry-wide benefits, it has also created a host of challenges in terms of reliability and control.

Percival is an intelligent companion that automatically detects 20+ failure modes—including incorrect tool use, context misunderstanding, and planning errors—while analyzing execution traces to identify long-term planning failures before they cascade into critical system breakdowns.

“AI agents are getting better at solving complex tasks, but their unpredictability presents serious challenges for developers and organizations,” said Anand Kannappan, CEO and Co-founder of Patronus AI. “When developers spend hours tracing through agent workflows only to find that a decision made five steps ago caused the final error, they’re not just losing time—they’re potentially losing control over their systems. Percival gives developers the ability to instantly understand and fix their AI agents, turning weeks of debugging into minutes while helping maintain essential human oversight as these systems grow more sophisticated.”

The platform leverages an agent-based architecture rather than a single LLM-as-judge model, enabling comprehensive error detection across four major categories:

• Reasoning Errors: including hallucinations, information processing, decision making, and output generation errors
• System Execution Errors: configuration, API issues, and resource management failures
• Planning and Coordination Errors: context management and task orchestration failures
• Domain Specific Errors: customized to specific workflow requirements

A key differentiator is Percival’s episodic memory system, which learns from previous errors and adapts to changing input distributions, making future error detection more reliable and customized to each organization’s workflow.

Unlike traditional evaluations for standalone LLMs, Percival addresses the unique challenges of agentic systems where early-stage decisions can manifest as errors in later pipeline stages. The platform maintains memory of previous failures, enabling customized benchmarking of agent systems.

Currently, AI engineers spend several hours per week debugging long agentic execution traces. Percival automates this process, reducing human effort required to analyze large agentic traces and accelerating development cycles.

Patronus AI’s vision of maintaining human oversight over AI workflows advances with Percival, representing a significant step toward reliable automated debugging of complex agentic systems.

“Emergence’s recent breakthrough—agents creating agents—marks a pivotal moment not only in the evolution of adaptive, self-generating systems, but also in how such systems are governed and scaled responsibly—which is precisely why we are collaborating with Patronus AI,” said Satya Nitta, Co-founder and CEO of Emergence AI. “While innovation remains at our core, we have always been equally committed to governance, transparency, and responsible deployment. Our collaboration strengthens that commitment by adding further depth to how we interpret, evaluate, and refine our agent-based systems. Together, we’re enhancing not just what’s possible, but how safely and responsibly it’s delivered at scale.”

For more information, visit https://www.patronus.ai/percival.

About Patronus AI

Patronus AI develops AI evaluation and optimization to help companies build top-tier AI products confidently. The company was founded by machine learning experts Anand Kannappan and Rebecca Qian. For more information, please visit https://www.patronus.ai/.

View original content:https://www.prnewswire.com/news-releases/patronus-ai-unveils-first-scalable-supervision-solution-for-agentic-systems-302455117.html

SOURCE Patronus AI