AI agents are everywhere, but they still stumble on complex work. Benchmarks show they often need close supervision: LiveBench 2.0 put Claude Fable 5.2 at 82.9 out of 100 tasks and ChatGPT-6 Astra at 75.6, while EON's simulated corporate test found Claude Opus 4.8 hitting 76.8% accuracy.

Plurai, a startup founded by Ilan Kadar and Elad Levi, wants to improve those numbers using methods borrowed from autonomous driving. The co-founders previously worked at Nexar, and Kadar helped Waymo simulate and evaluate its end-to-end deep learning driving software. "The moment that we saw the wave of AI agents, we saw the same pattern: That it is very easy to build a demo, but [hard] to get something that will be reliable and cost effective," Kadar told TechCrunch.

On Wednesday, Plurai said it raised $13 million in seed funding led by Team8, with participation from U&I, Velocity and Mercer. The platform builds a digital twin of a company's systems, maps the agent's possible flows, and uses that graph to prepare models with evaluations and fixes for edge cases. It works alongside tools like Claude Code, Codex, Copilot and Cursor.

Kadar says the bigger opportunity is post-training small, open-weight models tailored to each customer. Tests with Nvidia and Nebius showed Plurai's models beating mid-tier frontier models at lower cost. Angel investor and Google DeepMind researcher Reah Miyara said the edge comes from understanding production failure modes and generating synthetic data to tune models.

"My view is that no matter how smart OpenAI and Anthropic model will be - and they will be much smarter - they will not know your business and your policy," Kadar said.

Source: TechCrunch