Plurai raises $13M to test AI agents the way self-driving cars are tested

Started by Inland Aidan, Nov 17, 2026, 09:36 PM

Previous topic - Next topic

0 Members and 1 Guest are viewing this topic.

Topic: Plurai raises $13M to test AI agents the way self-driving cars are tested   Views(Read 56 times)
Active members in this topic:
Inland Aidan(1) NewDay(1)

Inland Aidan

A startup is borrowing techniques from autonomous vehicles to make AI agents more reliable. TechCrunch reports that Plurai has raised a $13 million seed round led by Team8, with U&I, Velocity and Mercer also investing

Co-founders Ilan Kadar and Elad Levi previously worked at Nexar, and Kadar worked with Waymo on simulation and evaluation. That background shapes their approach of simulating many scenarios before agents go live

The article cites benchmarks showing even top models falling short. On LiveBench 2.0, Claude Fable 5.2 succeeded on 82.9 of 100 tasks and ChatGPT-6 Astra on 75.6, while an EON study found Claude Opus 4.8 topping out at 76.8 percent

Kadar claims an order or two orders of magnitude cost reduction, based on tests with Nvidia and Nebius. Those performance claims are the company's own

Self driving cars are tested through millions of simulated miles, so applying the same idea to AI agents makes a lot of sense. Even the best models still fail a meaningful share of tasks, which shows why testing matters. Cheaper evaluation could help smaller teams build more reliable agents

How should AI agents be tested before they are trusted with real tasks? And what failure rate would you accept from an AI agent?

I read every reply. Even the bad ones.

NewDay

Even the best model failing 17 percent of tasks is sobering. Agents are impressive but not reliable enough for everything. Testing matters

Sobering numbers. Reliability lags