Plurai raises $13 million to improve the reliability of AI agents
AI agents still struggle with complex tasks without close supervision. In benchmark tests, Claude Fable 5.2 completed 82.9 of 100 tasks and ChatGPT-6 Astra completed 75.6, while Claude Opus 4.8 reached 76.8% accuracy in a simulated corporate environment evaluated by EON.
Plurai aims to improve those results through simulations modeled on techniques used in autonomous driving. The startup builds a digital twin of a customer’s systems, including its knowledge base and production tools. It maps possible agent workflows and uses the resulting graph to test and address difficult edge cases, such as ignoring travel spending rules, combining information from several databases, or losing track during long customer-service conversations.
The platform works with Claude Code, Codex, Copilot and Cursor. Plurai says its longer-term goal is to use the collected data to post-train smaller open-weight models for individual customers. Tests conducted with Nvidia and Nebius reportedly showed these models outperforming mid-tier frontier models at lower cost. The company announced a $13 million seed round led by Team8, with U&I, Velocity and Mercer also participating. Co-founders Ilan Kadar and Elad Levi previously worked at Nexar; Kadar also worked with Waymo on simulation and evaluation.
Written by Sakhanda Wire AI from reporting by TechCrunch. Facts, quotes and full context are in the original article below.