Program report · filed November 11, 2026
AI in SalesMeasured report
Plurai Raises $13M to Train AI Agents on Simulated Corporate Data
Plurai raised a $13M seed led by Team8 to simulate enterprise edge cases and post-train open-weight agents, after benchmarks showed top models under 83% task reliability.
By Rebecca Stone3 min read648 words
Program notes
- Plurai raised $13 million in seed funding on Wednesday, led by Team8 with U&I, Velocity and Mercer participating.
- EON's simulated corporate environment test found the best agent, Claude Opus 4.8, performed tasks at 76.8% accuracy.
- LiveBench 2.0 shows Claude Fable 5.2 and ChatGPT-6 Astra succeeding on 82.9 and 75.6 of 100 tasks.
- Co-founders Ilan Kadar and Elad Levi came from Nexar; Kadar simulated Waymo's end-to-end driving software.
- Tests with Nvidia and Nebius, per the company, showed Plurai's open-weight models beating mid-tier frontier models at one to two orders of magnitude lower cost.
The most capable AI agent tested in a simulated corporate environment completed tasks at 76.8% accuracy — a number too low for unsupervised deployment. Plurai, a startup founded by two veterans of autonomous driving company Nexar, raised $13 million in seed funding on Wednesday to raise those figures using the technique that put self-driving cars on the road: simulation on synthetic data.
Team8 led the round, with participation from U&I, Velocity and Mercer. Co-founders Ilan Kadar and Elad Levi previously worked at Nexar, where Kadar worked with Waymo to simulate and evaluate its end-to-end deep learning driving software.
How bad is the reliability problem?
Public benchmarks quantify the gap between vendor claims and measured performance. LiveBench 2.0 shows Claude Fable 5.2 and ChatGPT-6 Astra succeeding on 82.9 and 75.6 out of 100 tasks, respectively. A separate effort by AI infrastructure company EON, which tested agents in a simulated corporate environment, found the best performer — Claude Opus 4.8 — reached 76.8% task accuracy.
The failure patterns are concrete. Agents asked to book travel can ignore company spending policies when pressed. An agent that reliably queries a single database breaks down when it must combine several data sets. Long customer service conversations overload the agent with context and push it toward wrong answers.
"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. "Since we saw the same pattern, the same pitch was that we want to build this simulation layer."
What does the simulation layer actually do?
Plurai builds a digital twin of a company's electronic systems — the knowledge base and the tools an agent will use in production. It then maps every possible agent flow across that twin and uses the resulting graph to run evaluations and prepare fixes for edge cases before deployment. The platform runs alongside existing agentic tools including Claude Code, Codex, Copilot and Cursor.
Kadar frames the larger prize as post-training. Plurai gathers data from the simulations and uses it to train small, open-weight models tailored to a single customer's needs. Tests run jointly with Nvidia and Nebius, according to the company, showed Plurai's models beating mid-tier frontier models at lower cost. This is a vendor-reported result without disclosed methodology or sample size.
"We can take a large language model, we can take the task, decompose it into a smaller task using the simulation, train [an open-weight] model, and using the simulation, verify that the quality is higher or comparable to the frontier model, and an order or two orders of magnitude cost reduction," Kadar said.
Why investors back the cost argument
Reah Miyara, a Google DeepMind researcher who backed Plurai as an angel investor, said the company's edge lies in understanding failure modes in production and generating synthetic data to tune models against them. She pointed to what IT leaders are telling their teams: "frontier-capable models are out of distribution with my specific data, and the cost of these things isn't sustainable in the long term."
The startup draws its forecast from robotics. General-purpose AI models give robots impressive capabilities, yet those robots remain too unreliable for wide deployment. Meanwhile, robots running task-specific AI models already work for customers in the field and collect real-world data. Plurai expects enterprise agents to follow the same split.
"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.
The $13 million now funds the bet that simulated edge cases, not larger frontier models, will decide which agents enterprises can actually trust.
via techcrunch.com (Source)
More from Rebecca Stone
Show full bio
Market editor covering people, launches and day-to-day practice in Sales technique and methodology, sales enablement at Quota Brief.
139 articles
Also rated
- Software Firms Ship AI Sales Agents 4x Faster Than Other Industries
- FDE Hiring Up 1,000%: The Human Cost Inside 'Autonomous' AI
- Gartner's 2026 CRM CEC Quadrant: Five Leaders, One Battle — Governed AI Autonomy
- 77% of Enterprises Reevaluate AI Vendors Every Six Months
- Software Firms Ship AI Sales Agents Four Times Faster Than Peers