Plurai wants to help AI agents work better using the same approach that helped get self-driving cars to roads
Like the end-to-end models trained to drive cars, AI agents are probabilistic systems operating in unpredictable environments.
Plurai, a startup founded by Ilan Kadar and Elad Levi, aims to enhance AI agent performance by employing a model similar to what has enabled self-driving cars to operate on roads. Currently, AI agents struggle with reliability and cost-effectiveness when handling complex tasks, as evidenced by benchmark tests where Claude Fable 5.2 and ChatGPT-6 Astra achieved 82.9% and 75.6% success on 100 tasks, respectively.
EON, an AI infrastructure company, found Claude Opus 4.8 successful in a simulated corporate environment, with a 76.8% accuracy rate.
To address these issues, Plurai develops digital twins of a company's electronic systems and maps the possible agent flows using these graphs. This simulation layer allows for evaluations and fixes of potential edge cases, much like the process used to put self-driving cars on the road. The startup's tools integrate seamlessly with existing agentic tools like Claude Code, Codex, Copilot, and Cursor.
Plurai's platform generates synthetic data by decomposing tasks into smaller tasks using simulation, training open-weight models for individual customer needs, and verifying their quality against mid-tier frontier models while significantly reducing costs. According to Reah Miyara, a Google DeepMind researcher who supported Plurai as an angel investor, the company's edge lies in its ability to understand production failure modes and generate synthetic data to optimize model quality.
As AI agents continue to advance in both digital and physical environments, Plurai believes that their approach could prove advantageous for businesses seeking to optimize AI expenses and improve agent reliability.
Written by urgent.news from TechCrunch's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
