Plurai raises $13M to train AI agents like self-driving cars
Ex-Nexar engineers Ilan Kadar and Elad Levi have raised $13 million to bring self-driving car simulations to the enterprise AI agent market. Their startup, Plurai, builds digital twins of corporate systems so AI models can train on synthetic data before tackling real tasks. The seed round was led by Team8.
Editor, Lazyfounder
30 SEC SUMMARY
- Plurai, founded by ex-Nexar engineers, raises $13M to train AI agents using self-driving car simulation tech.
- The startup builds digital twins of corporate systems to test AI models for edge cases—before real-world deployment.
- Plurai’s models outperformed mid-tier AI rivals in tests with Nvidia and Nebius, cutting costs by 10-100x.
- Its approach mirrors how autonomous vehicles rely on synthetic data to refine AI before hitting the road.
- The seed round, led by Team8, targets enterprise AI reliability gaps left by general-purpose models.
TABLE OF CONTENTS
$13 million seed round takes simulation from cars to cubicles
Plurai, a startup built by two ex-Nexar engineers, has closed a $13 million seed round. Team8 led the investment, with cash also coming from U&I, Velocity, Mercer and Google DeepMind researcher Reah Miyara.
The money will fund a platform that trains enterprise AI agents inside simulated copies of a company’s own systems. By stress-testing models on synthetic data, Plurai aims to catch errors before they reach production.
Co-founders Ilan Kadar and Elad Levi cut their teeth at Nexar, a company that builds autonomous-driving tech, and Kadar later worked on simulation at Waymo.
How self-driving car tech trains enterprise AI
Kadar and Levi are applying the simulation playbook they wrote for self-driving cars to the enterprise AI agent boom. At Nexar and Waymo, they used digital twins of streets to train and evaluate autonomous vehicles. Now Plurai creates digital twins of corporate IT stacks—databases, APIs, policy engines—to prepare AI models for the chaos of real work.
The result is models that can handle edge cases without constant human hand-holding. A recent TechCrunch analysis found today’s AI agents succeed on only 75-83 out of 100 tasks; Plurai believes its simulations can close that gap by exposing models to the messiest scenarios before they ever touch live data.
Why benchmarks lie about AI agent reliability
AI agents built on models like Anthropic’s Claude Opus 4.8 or OpenAI’s ChatGPT-6 Astra hit 75-83% accuracy in benchmark tests, according to TechCrunch. But those tests often ignore the fluidity of real corporate environments—shifting data, unexpected user behavior, strict compliance rules.
To fix the mismatch, Plurai generates synthetic data that mirrors a customer’s actual systems. Models train on thousands of simulated failures until they master tasks without crashing. In tests with Nvidia and Nebius, Plurai’s models outperformed mid-tier frontier models while costing 10–100× less.
Cheaper, smaller models outperform generic giants
Instead of selling another general-purpose model, Plurai focuses on small, open-weight models fine-tuned for a single customer. These models slot into existing agentic stacks like Claude Code, Copilot or Cursor and learn from live interactions. Reah Miyara, an early investor, argues that most companies simply can’t shoehorn a big frontier model into their unique workflows—Plurai’s tailored models fill that gap.
Miyara also points out that the market already tolerates task-specific AI in robotics. Factories use robots with bespoke vision models; Plurai is betting the same playbook works for enterprise software.
What autonomous vehicles teach AI agents
Self-driving cars and AI agents share a core problem: both operate in unpredictable environments where mistakes cost real money. Simulation is the cheapest way to rehearse the dangerous stuff—whether it’s a Tesla avoiding a pedestrian or a customer-service bot canceling a contract.
While OpenAI and Anthropic chase ever-larger models, Plurai is following the robotics playbook: small, specialized models that do one thing well. Early results suggest the strategy pays off; if it scales, the AI-agent market could look very different a year from now.
What this means
Lazyfounder analysis — our interpretation, not reported fact.
Plurai’s digital-twin method lets enterprises stress-test AI agents in private simulations—avoiding the reputational and financial risks of public failures. If its small, custom models keep beating mid-tier giants at a fraction of the cost, expect OpenAI and Anthropic to take notice.
Key takeaways
- Digital twins built from real corporate data are now being used to harden enterprise AI agents before live deployment.
- Simulation-driven training can slash AI costs by 10-100× compared to general-purpose models—without sacrificing accuracy.
- Ex-Waymo engineers are applying autonomous-vehicle simulation playbooks to the AI-agent boom.
- Early enterprise wins could force a reckoning for the ‘bigger-is-better’ philosophy of frontier AI labs.
Related on Lazyfounder
Sources
- TechCrunch · 2026-11-11
Plurai wants to help AI agents work better using the same approach that helped get self-driving cars to roads
This story is an original summary drafted with AI by Lazyfounder from the reporting listed above and checked by automated validation. Facts are attributed to their original publishers; sections marked as analysis are Lazyfounder's. Where a source is in another language, facts were machine-translated and quotations are reported, not reproduced. Read the original coverage via the links, and see our AI policy and corrections policy.
About the author
Editor, Lazyfounder
Tarun Mottlia edits LazyFounders, covering Indian startups, funding rounds, AI and product launches. Every story on the site is AI-assisted and checked against its cited sources before publication.
More stories by Tarun MottliaGet the LazyFounder Brief
Startup, funding and AI news in a five-minute read. Join the early-access list.
