Today we’re launching Plurai simulation, the self-improving layer for AI agents. It’s free to start for every AI engineer, and we’re announcing $13M in seed funding led by Team8. The story behind it starts on the road.
Elad and I spent years building AI for autonomous driving, working with companies like Waymo. In that world, mistakes aren’t an option and simulation is how you earn trust. We learned that AI doesn’t fail on the happy path. It fails in the long tail, in situations no one predicted. Waymo earned trust with more than 20 billion simulated miles, roughly 100 times its real-world driving. The rarest events were met, and fixed, in simulation first.
When we started working with teams building AI agents, it felt like déjà vu. They had brilliant demos, and then real customers arrived with the long tail. Months of manual checking, a launch everyone felt good about, and then a failure no one saw coming. And going live only raised harder questions. How do you catch failures in production and keep improving the agent as its tools, data and users change, without endless rounds of manual review and testing? And how do you scale at all, when frontier models are expensive to run?
So we founded Plurai to build the recursive self-improvement layer for AI agents, starting with simulation. It creates a digital twin of an agent and its world from the company’s own knowledge base, tools and policies. Then it fills a sandbox with realistic users, covering the full range of intents, personas and edge cases the agent will meet in the real world, exposing policy violations, hallucinations, incorrect actions, broken tool calls, failed tasks and sensitive data leaks before any customer sees them. But finding failures was never enough. For every failure, the simulation explains why it happened and generates the fix, so developers stop debugging by hand. Plurai also uses that same simulation data to post-train models tailored to the agent, making it more capable while cutting the cost of running it at scale. In production, every new failure feeds back into the simulation, so the agent keeps getting better: simulate, fix, post-train, repeat. That’s what we mean by recursive self-improving agents.
Over the past months, we put this to work with a select group of design partners across AI-native and highly regulated industries, including Fortune 500 and major financial companies. Their agent development cycles went from months to hours. Failures reaching real users fell by 82%, and the cost of running reliable agents fell by 97%. In joint work with NVIDIA and Nebius, models Plurai post-trained for a global enterprise customer outperformed OpenAI and Anthropic frontier models on accuracy, at 40x and 44x lower cost, respectively.
That’s why today matters so much to me. The autonomous driving industry needed huge teams and huge budgets to build its simulators. AI agents shouldn’t need that. Starting today, any agent builder can use Plurai simulation for free, right from the tools they already use, including Claude Code, Codex and GitHub Copilot. Teams can also bring agents built on enterprise platforms like Salesforce Agentforce.
Our $13M seed round was led by Team8, with U&I, Velocity and Mercer, and more than 20 angels and industry leaders. They include CyberArk founder Udi Mokady and Reah Miyara, Senior Director of Google Models at Google Cloud and former model evaluation lead at OpenAI. Dr. Eyal Carmi, CTO of Microsoft 365 Copilot, has joined our advisory board. With their support, we’re just getting started.
To our design partners: you trusted us early and pushed us hard, and today exists because of you. To Team8 and everyone who backed us: thank you for believing in this and joining our journey before it was obvious. And to the amazing Plurai team: I’m so proud of what we’ve built together. Watching you push the boundaries of what’s possible every single day is the greatest privilege of my career. Thank you, from the bottom of my heart.
Self-driving cars earned our trust one simulated mile at a time. Now it’s the agents’ turn. If you’re building one, start for free with Plurai simulation and tell me what you think: simulation.plurai.ai/connect
Ilan
Co-Founder & CEO, Plurai
