Plurai raises $13M to fix AI agents with self-driving tech
Original: Plurai wants to help AI agents work better using the same approach that helped get self-driving cars to roads
Why This Matters
Agent reliability is the key bottleneck blocking enterprise AI adoption at scale.
Plurai, a startup using autonomous-driving simulation techniques, raised $13M in seed funding to make AI agents more reliable. Led by Team8, the round aims to build synthetic-data training for enterprise agent deployments.
AI agents are underperforming. LiveBench 2.0 scores Claude Fable 5.2 at 82.9/100 and ChatGPT-6 Astra at 75.6/100, while EON's simulated corporate environment found even the top model, Claude Opus 4.8, hitting only 76.8% task accuracy — not nearly good enough for unsupervised enterprise use.
Plurai's answer borrows directly from autonomous vehicle development. Co-founders Ilan Kadar and Elad Levi come from Nexar, an AV technology firm; Kadar also worked with Waymo on simulating and evaluating its deep learning driving software. Their thesis: agents fail for the same reasons early self-driving systems did — edge cases and unpredictable environments.
The platform builds a digital twin of a company's internal systems, maps all possible agent flows, then uses that graph to surface and patch edge cases before deployment. It sits alongside existing tools like Claude Code, Codex, Copilot, and Cursor.
The bigger play, though, is post-training. Plurai uses data gathered from simulations to fine-tune small, open-weight models for specific customer needs. Joint tests with Nvidia and Nebius showed these models matching or beating mid-tier frontier models at significantly lower cost. The $13M seed round was led by Team8, with U&I, Velocity, and Mercer participating.