# Snorkel AI Boosts Multiagent Evaluation Accuracy and Efficiency > With the integration of Prisma AIRS AI Gateway, Snorkel AI transforms its debugging workflow, gaining unprecedented visibility into multi-agent evaluations. This leap forward not only streamlines operations but also enhances the quality of outputs. **Source**: paloaltonetworks.com | **Published**: 2026-11-02 | **Type**: case_study ## Key Facts - Snorkel AI achieved a 20% accuracy boost, enhancing competitive edge in agent evaluations. - AI Gateway enables 2x faster issue detection, reducing time-to-market for new features significantly. - Visual execution paths reveal hidden vulnerabilities, improving debugging efficiency and reliability. - Faster iteration cycles cut development costs, allowing for more resources to be allocated to innovation. - Enhanced visibility into agent performance indicates a strategic shift towards data-driven decision-making. ## Summary \## Summary Snorkel AI faced challenges in debugging their multi-agent evaluation workflows, which involved processing thousands of results weekly. They implemented Prisma AIRS AI Gateway to enhance visibility into agent execution paths, resulting in a 20% accuracy boost in evaluations and a significant reduction in issue detection time. \## Background Snorkel AI operates in the artificial intelligence sector, focusing on evaluation workflows that review expert-generated labels for specialized domains. Prior to deploying AI Gateway, the team struggled with a fragmented debugging process, relying on scattered logs and manual reconstruction of agent execution paths. \## Challenge The primary challenge was the inefficiency in identifying and resolving issues within their multi-agent evaluation system. The existing debugging workflow was cumbersome, making it difficult to trace errors and understand the decision-making process of agents. \## Solution The team implemented Prisma AIRS AI Gateway, which provided a visual hierarchy of agent calls, subagent interactions, and tool invocations displayed as a nested tree. This allowed for individual prompt-response inspection at any node and enabled custom tagging and filtering to identify patterns across traces. Side-by-side trace comparisons facilitated understanding of divergences between successful and failing executions. \## Results The deployment of AI Gateway led to measurable improvements in Snorkel AI's evaluation workflows: - A 20% increase in accuracy for agent evaluations, enabling the team to identify and address previously hidden edge cases. - A twofold increase in the speed of issue detection during development, with problems surfacing immediately in traces instead of requiring extensive log analysis. - A significant reduction in iteration cycles, with tasks that previously took hours of manual work now completed in minutes through visual inspection. \## Key Insights Implementing a visual debugging tool can drastically improve the efficiency of identifying and resolving issues in complex systems. Enhanced visibility into execution paths allows teams to catch errors earlier and iterate more quickly. \## Customer Testimonial No direct quote is available from the source material. ## Entities - **Companies**: Snorkel AI, Palo Alto Networks - **Products**: Prisma AIRS AI Gateway - **Technologies**: Reinforcement Learning from Human Feedback (RLHF) ## Key Concepts Multi-Agent Question-Answer Validator, debugging workflows, execution paths, nested tree visualization, custom tagging and filtering, trace comparison, accuracy boost, issue detection ## Definitions - **Multi-Agent Question-Answer Validator**: A tool used by Snorkel AI to process and validate thousands of results weekly for expert-generated labels. - **Reinforcement Learning from Human Feedback (RLHF)**: A machine learning approach that uses feedback from human evaluations to improve the performance of AI models. - **AI Gateway**: A feature of Prisma AIRS that provides visual hierarchy and insights into agent execution paths. - **debugging workflow**: The process of identifying and resolving issues within software systems, often involving the analysis of logs and execution paths. - **aggregate metrics**: Summary statistics that provide an overview of performance but may obscure individual case details. ## Use Cases - Improving accuracy in agent evaluations - Faster issue detection during development - Visual inspection of execution paths - Identifying patterns across traces - Comparing successful and failing execution paths ## Frequently Asked Questions **What is the main benefit of using Prisma AIRS AI Gateway?** The main benefit is the enhanced visibility it provides into agent execution, allowing teams to quickly identify and resolve issues that were previously hidden in logs. **How does AI Gateway improve debugging workflows?** AI Gateway improves debugging by visualizing execution paths as nested trees, enabling easier inspection of individual calls and comparisons between successful and failing traces. **What kind of improvements did Snorkel AI see after implementing AI Gateway?** Snorkel AI experienced a 20% boost in accuracy for agent evaluations and doubled the speed of issue detection, significantly reducing the time needed for manual log analysis. **Can AI Gateway help in identifying edge cases?** Yes, AI Gateway allows teams to catch and fix edge cases that were previously invisible in aggregate metrics, leading to more robust evaluation workflows. **What is the significance of visualizing execution paths?** Visualizing execution paths helps teams understand the flow of operations within agents, making it easier to pinpoint where things go wrong and facilitating faster troubleshooting. ## Links - [Read on Welcome.AI](https://welcome.ai/content/snorkel-ai-boosts-multiagent-evaluation-accuracy-and-efficiency) - [Original source](https://www.paloaltonetworks.com/customers/how-snorkel-ai-debugs-multi-agent-evaluations-at-scale-with-ai-gateway) - [Palo Alto Networks](https://welcome.ai/company/palo-alto-networks): Featured company --- Source: Welcome.AI | https://welcome.ai/content/snorkel-ai-boosts-multiagent-evaluation-accuracy-and-efficiency