Plurai raises $13M to crash-test AI agents the way Waymo tests cars

Plurai launched Plurai simulation on Nov. 11, 2026, a free-to-start tool that stress-tests AI agents against simulated users and edge cases, and said it raised a $13 million seed round led by Team8.

Everyone has an agent demo; far fewer have an agent they'd let near a real customer, and the gap is all the weird stuff nobody tested. Plurai is betting the self-driving playbook, simulate the long tail before it happens, transfers to software agents. The 82% and 97% numbers are the company's own, from hand-picked design partners, so the real proof is what happens now that anyone can try it free.

Plurai said it raised a $13 million seed round led by Team8. A seed is the early private check, before a larger round. U&I, Velocity, and Mercer also took part, plus more than 20 angels. Angels are people investing their own money. Named among them: CyberArk founder Udi Mokady, and Reah Miyara, senior director of Google Models at Google Cloud, who previously led model evaluation at OpenAI. Dr. Eyal Carmi, chief technology officer of Microsoft 365 Copilot, joined the advisory board. The announcement is dated Nov. 11, 2026. The amount and the names are Plurai's, in a post by co-founder and chief executive Ilan Kadar. TechCrunch reported the same round the same day.

Co-founders Ilan Kadar and Elad spent years building AI for autonomous driving, working with companies like Waymo, Kadar wrote. The post gives Elad's first name only. That background is Kadar's.

The product, Plurai simulation, builds a digital twin of an agent and the world that agent works in. A digital twin, here, is a software copy of that setup, drawn from a company's own knowledge base, its tools, and its policies. The tool then fills a sandbox with simulated users. A sandbox is a closed practice room, not the live product. Those users cover different intents, personas, and edge cases. An edge case is the odd situation almost nobody tries until a customer hits it. The aim is to expose policy violations, hallucinations, wrong actions, broken tool calls, failed tasks, and data leaks before a person does. A hallucination, here, is the agent stating something that is not true. For each failure, Plurai says, the tool explains why and generates a fix. Those lines are the company's.

Plurai also says it uses that simulation data to post-train smaller models tailored to the agent. Post-train means extra training after a model already exists, aimed at this one job. Failures from the real product feed back into the simulation. That loop is the company's description.

Plurai says the simulation is free to start for every AI engineer. The launch post is at https://www.plurai.ai/blog/introducing-plurai-simulation-self-improving-layer-for-ai-agents. It can be used from Claude Code, Codex, and GitHub Copilot. TechCrunch also lists Cursor. Companies can bring agents built on Salesforce Agentforce. The free-to-start line and the tool list, aside from Cursor, are Plurai's. Cursor is TechCrunch's addition.

Plurai's own claims, from design partners it describes as Fortune 500 companies and major financial companies: development cycles went from months to hours, 82 percent fewer failures reaching users, and 97 percent lower cost. In work with NVIDIA and Nebius, models it post-trained beat OpenAI and Anthropic frontier models on accuracy, at 40 times and 44 times lower cost. A frontier model is one of the most capable general models from those labs. These figures are the company's, from hand-picked design partners. Plurai did not attach an outside audit.

Kadar's analogy is Waymo. He wrote that Waymo used more than 20 billion simulated miles, about 100 times its real-world driving. That comparison is his. Waymo is the self-driving car company. The point he is making is that the cars learned the rare cases in simulation, not only on the road.

The picture is the Plurai team, about 15 people in matching black Plurai T-shirts, posing outdoors between trees and office buildings, with the front row crouching. The credit is Guy Hamoy / Plurai (via TechCrunch).

In plain terms, a startup whose founders came from self-driving cars raised $13 million, led by Team8, and opened a tool that crash-tests an AI agent against simulated users before a customer does. It is free to start. The 82 percent and 97 percent figures, and the claim of beating frontier models for far less money, are Plurai's, from its design partners. The round does not come with a valuation or a revenue number.