Plurai Raises $13M to Make AI Agents More Reliable

AI agents are becoming increasingly popular, but getting them to perform complex business tasks without human supervision remains a major challenge. Startup Plurai believes the solution could come from a technology that helped advance self-driving cars: simulation and synthetic data.

The company is applying techniques developed in the autonomous driving industry to make AI agents more reliable, accurate, and cost-effective for businesses. On Wednesday, Plurai announced that it had raised $13 million in seed funding to develop its platform.

The funding round was led by Team8, with participation from U&I, Velocity, and Mercer.

AI Agents Still Struggle With Complex Tasks

Despite rapid progress in artificial intelligence, current AI agents are not always dependable when handling complicated workflows.

Recent benchmarking efforts highlight the problem. LiveBench 2.0 found that Claude Fable 5.2 and ChatGPT-6 Astra successfully completed 82.9 and 75.6 tasks out of 100, respectively. Separately, AI infrastructure company EON tested agents in a simulated corporate environment and found that Claude Opus 4.8, the strongest model in its evaluation, achieved 76.8% accuracy.

These results suggest that AI agents still need significant human oversight before companies can confidently rely on them for important operations.

For example, an AI agent responsible for booking business travel might ignore company spending limits. Another agent could successfully retrieve information from one database but struggle when asked to combine information from several different sources. Similarly, lengthy customer service conversations can overload an agent with context, increasing the chances of mistakes.

Plurai wants to address these problems before AI agents are deployed in real business environments.

Founders Bring Self-Driving Car Experience to AI

Plurai was founded by Ilan Kadar and Elad Levi, who previously worked at Nexar, a company focused on autonomous driving technology.

Kadar also worked with Waymo on simulating and evaluating its end-to-end deep learning driving software.

Their experience in autonomous vehicles helped them recognize a similar challenge emerging in the AI agent industry.

According to Kadar, building an impressive AI agent demonstration is relatively easy, but making the technology reliable and affordable enough for everyday business use is much harder.

He explained that the team recognized the same pattern in AI agents that they had previously encountered in self-driving technology. This led them to develop a simulation layer designed to improve how agents perform under difficult conditions.

How Plurai’s Simulation Platform Works

Plurai creates a digital twin of a company’s digital infrastructure, including its internal knowledge base and the tools that AI agents will use once deployed.

The platform then maps the different workflows an agent might encounter and builds a graph of possible interactions. Using this information, it runs evaluations and prepares agents to handle unusual situations, known as edge cases.

The goal is to identify potential failures before they affect actual business operations.

Plurai’s platform is also designed to work alongside existing AI development tools, including Claude Code, Codex, Copilot, and Cursor.

Rather than simply testing whether an agent can complete a task, the company wants to understand where it might fail and use those findings to improve its performance.

Training Smaller AI Models to Reduce Costs

Beyond simulation and evaluation, Plurai sees another major opportunity in using the data it collects to train smaller, open-weight AI models for individual businesses.

The company believes businesses do not always need expensive, large-scale frontier models to handle specialized tasks.

Kadar said Plurai can take a large language model, break a complicated task into smaller components through simulation, and use the resulting data to train an open-weight model. The platform can then evaluate whether the smaller model delivers comparable or better results.

Tests conducted in cooperation with Nvidia and Nebius showed that Plurai’s models outperformed mid-tier frontier models while operating at lower costs.

Kadar said this approach could reduce expenses by one or even two orders of magnitude while maintaining comparable performance.

Why Investors See an Opportunity

Reah Miyara, a Google DeepMind researcher and angel investor in Plurai, believes the startup’s strength comes from understanding how AI systems fail in real-world production environments.

By generating synthetic data based on those failures, Plurai can help businesses customize and improve their AI models.

Miyara also pointed to growing concerns among IT leaders about the long-term cost of using advanced AI models. General-purpose frontier models may perform well in broad benchmarks but struggle with company-specific information, workflows, and policies. Their operating costs can also become difficult to sustain.

Plurai believes its approach could help companies get more value from AI without depending entirely on expensive frontier models.

The Future of Business AI Agents

Plurai’s strategy is based on a comparison between AI agents working in digital environments and AI-powered systems operating in the physical world.

Although advanced general-purpose AI models are giving robots impressive capabilities, they are not yet reliable enough for widespread deployment in many situations. Meanwhile, robots designed for specific tasks are already working in real-world environments, serving customers and collecting valuable operational data.

The startup believes a similar approach could work for business AI agents: Instead of relying only on increasingly powerful general-purpose models, companies could develop specialized systems trained around their own workflows and requirements.

Kadar argues that even as OpenAI and Anthropic continue improving their AI models, those systems will not automatically understand every company’s internal processes, business rules, or policies.

For Plurai, that gap represents a significant opportunity. By combining simulation, synthetic data, and specialized model training, the company wants to help businesses move AI agents beyond impressive demonstrations and toward reliable everyday operations.

Also read : Drops Music App Brings Social Music Discovery Back to Life