Patronus AI debuts Percival to help enterprises monitor failing AI agents at scale

Batronus ai It has launched a new monitoring platform that automatically automatically defines the failures in the AI ​​Agent systems in the company, which are reliably growing up to the complexity.

New product of San Francisco based AI security start, PassengerAs the first solution to the AI ​​Agent systems, the first solution to identify various samples automatically, itself offers itself to optimize them to solve them.

“Percival, agent systems automatically identify various samples of failure and then solve adjustments and optimizations in systematically,” he said.

AI Agent Reliability Crisis: Why companies lose control of autonomous systems

Acceptance of AI agents-software that can be independent of independent complex tasks independentlyaccelerated In recent months, it is trying to create new management problems such as companies, ensure that these systems are valid on a scale.

Unlike ordinary machine learning models, this agent-based systems often attract the long sequence of operations where the result of the low flow of errors in early stages.

“A few weeks ago, we published a model that agents could fail and what could affect the brand, client bar and such things.” Cannappan. “There is a permanent complicated error with agents we have seen.”

This issue is especially sharp in multifent environments where various AI systems interact with each other are becoming increasingly insufficient approaches of the traditional test.

Episodic memory innovation: Percival AI agent revolutionizes error detection

Passenger Distinguishes herself from other assessments with the ability to name agent-based architecture and the company called “episodic memory” and adapt to special workflows.

The program can detect more than 20 different failures in four categories: justification errors, system performance errors, planning and coordination errors and domain special errors.

“Unlike an LLM like a judge, it is an agent, and therefore can watch all the events that occur throughout the trajectory,” he said. “It can coordinate them and finds these mistakes in contexts.”

For businesses, the nearest benefit has been reduced. According to Patronus, early clients have reduced the time analyzing Agene’s workflows from one hour from an hour to another from an hour.

Trail Benchmark AI reveals critical gaps in control capabilities

Along with the crop launches, leaving a criterion called Patronus Trail (Trace and Agency Issue Localization) AI Agent to evaluate how the problems can detect the workflows.

Use This criterion Even complex AI models also hit only 11% compared to the best performing system, which fights effective trace analysis.

Findings emphasize the difficult essence of monitoring monitoring of the monitoring complex, and why large enterprises can help invest in specializing in specialized vehicles for AI.

Enterprise AI leaders receive Percival for Mission-Critical Agent Applications

Includes early elevators Emerging aiIt is raised nearly Funded $ 100 million And the systems can create and manage other agents of AI agents.

“Agents’ latest leash agents, not only in the evolution of adaptation, not the evolution of self-created systems, but also assessed how such systems evaluate and the results of the founder, said the founder AI said.

Another early customer uses a platform technology that helps large enterprises transfer the inheritory code through AI-powered SAP integrations.

These customers aim to solve the problem of the problem. According to Kannappan, some companies manage agent systems that are “100 steps in an agent folder” that produces complexity that can effectively control human operators.

The AI ​​control market was prepared to grow explosive as the spread of autonomous systems

Starting comes between an increasing enterprise in connection with the AI ​​reliability and management. Because companies have increased autonomous systems, the need for control instruments increased in proportion.

“How difficult these systems are increasingly,” Kannappan said that using the AI, using AI, using AI, “using AI,” the establishment of an environment in favorable order.

AI monitoring and reliability tools are expected to be significantly expanded as the experimental deployments of enterprises pass through the experimental deployments of the mission-critical AI.

Percival integrates face to face, including a large number of AI frames Sulolents, Pdyacik, Openai Agent SDKand LangchainIt adapts it with different development environments.

While Batronus ai Disclosures the price or income forecasts, direction of the company’s enterprise-grade control center, the forecast of these analysts shows that the analytics forecasting the AI ​​will increase significantly.