Plurai raises $13M to make AI agents more reliable through simulation

AI agents are everywhere in pitch decks. In production, they are still the coworker you cannot leave alone with the client.

Plurai, a startup building what it calls a reliability platform for AI agents, has raised $13 million, according to TechCrunch. The company’s thesis is that agents fail in the wild because nobody tested them against enough of the wild first.

The reliability gap Plurai is betting on

The problem is easy to state. Agents asked to handle complex, multi-step work generally are not dependable enough to run without a human watching closely.

TechCrunch’s reporting points to benchmarking efforts on complex agent tasks, referencing an 8% accuracy figure. That is not the performance profile of a system you hand the keys to.

Plurai’s answer is simulation. Its engine throws thousands of synthetic, multi-turn scenarios at an agent to see where it breaks before a customer finds out.

The company also uses a process it calls “vibe-training.” Teams describe the rules an agent should follow in plain language, and the platform turns those descriptions into custom small language models, or SLMs.

SLMs are compact AI models built for narrow jobs. Instead of asking a giant general-purpose model to police itself, a smaller specialist is trained to enforce specific policies, which tends to be faster and cheaper.

The numbers behind the pitch

Plurai makes some aggressive performance claims, and they are worth reading as company figures rather than independent results.

The startup says its approach produces over 43% fewer evaluation failures than GPT-5-mini, the OpenAI model it uses as a benchmark. It also claims its platform runs at 8 times lower cost.

On speed, Plurai cites sub-100ms latency for real-time guardrails. Put simply, the safety checks run fast enough to sit inside a live conversation without making users wait.

The company further claims up to 15 times greater edge-case coverage through its simulation techniques.

Research compiled on the company also points to an approximately $10 million seed or early-stage round dated February 1, 2026. Team8, U&I Ventures and Mercer Ventures participated in that round.

Who is behind it

Plurai was founded by Dr. Ilan Kadar, who serves as CEO, and Dr. Elad Levi, the CTO. Both have backgrounds in autonomous systems.

Kadar holds a PhD in machine learning and computer vision. Levi holds a PhD in mathematics.

The company is based in New York and has significant ties to Tel Aviv.

Plurai has also released IntellAgent, an open-source framework for multi-agent simulation aimed at evaluating conversational AI. It came out in the 2025 to 2026 window.

The startup also has strategic partnerships with NVIDIA, working with the chipmaker’s Nemotron models and NIM software. NIM is NVIDIA’s packaging for deploying AI models as ready-to-run services.

What this means for the agent economy

There are real questions to watch. The 43% failure reduction, the 8 times cost advantage and the 15 times edge-case coverage are company claims, and enterprise buyers will want to see them hold up across their own messy, specific workflows.

Synthetic scenarios also have a known limitation. A simulation is only as good as the imagination of whoever designs it, and real users have a long track record of finding failure modes nobody thought to test.

The next milestone to watch is adoption. Downloads of IntellAgent, production deployments of vibe-trained models and named enterprise customers will indicate whether Plurai’s simulation-first approach becomes standard practice or remains a niche testing tool.