Corvic Labs has launched, aiming to provide open operational infrastructure for AI developers and researchers. The initiative focuses on creating services and products to evaluate and govern agentic AI systems, addressing a critical industry need as AI agents become more prevalent in multi-step, autonomous tasks.
"Enterprises have difficulty launching products into production; these systems could be wild beasts," stated Corvic Chief Executive Farshid Sabet. He emphasized the importance of giving customers confidence in deploying AI products.
Corvic Labs will operate independently from Corvic AI's commercial platform, maintaining neutrality and focusing on open, free developer tooling. The initial release is the Agentic MCP Evaluator, a platform designed to simplify testing and evaluation of multistep AI agents. This tool allows developers to attach an evaluation framework via Anthropic PBC's Model Context Protocol, enabling connection to other AI models and third-party sources.
The evaluator aims to address challenges enterprises face in reproducing AI hallucinations and understanding accuracy. Sabet highlighted that measuring AI success is often subjective and not easily repeatable. Corvic Labs' approach utilizes deterministic workflows and domain metrics to make measurements systematic and comparable.
As AI product teams adjust data or models, agentic behavior can change unpredictably. The Agentic MCP Evaluator is designed to help teams manage these changes, test reliability, reasoning quality, and tool use in a standardized and democratized manner.