Every tech ecosystem has its defining archetypes. In Israel, one of them is the founding engineer: the person who joins a startup before the first line of code, before product-market fit and before anyone is sure the idea will work. It is a role filled with intensity, uncertainty and responsibility. Few people embody that journey better than Raz Rotenberg, co-founder and CEO of Fabrix.
Before starting Fabrix, Raz spent five years as the founding engineer at Run:AI, helping shape the company from its earliest technical decisions to its eventual acquisition by Nvidia. His new company, Fabrix, is now building AI-driven identity security at a moment when security teams are overwhelmed and traditional tools are not keeping up.
The story behind Fabrix reveals how Israeli engineering culture, AI breakthroughs and personal conviction combine to create the next wave of cybersecurity startups.
What It Really Means to Be a Founding Engineer
People often imagine founding engineers as simply early developers. Raz explains that the reality is different. The job is half engineering and half cultural DNA.
When he joined Run:AI, nothing existed except an idea and a small team. The founding engineer writes the first architecture, makes hundreds of micro decisions that shape the future product and carries the emotional weight of early failures and early wins.
“You bring your attitude, your energy, your DNA. That is what the company will be built on,” he says.
And the hardest part is not the technology. It is the rejection.
Proof-of-concept pilots fail. Deals fall apart. Customers decline to continue. In a mature company, those moments are buffered by scale. In a newborn startup, they feel personal.
Raz still remembers the logos of the companies that rejected Run:AI’s early POCs. They stay with him.
But the highs stay too. One of his clearest memories is the evening he tested a new idea on his laptop that later became fractional GPU, one of Run:AI’s defining features. At 9 p.m. it simply worked. He called the founders immediately.
“That spark became one of the core features of the platform,” Raz recalls.
Inside the Run:AI Acquisition by Nvidia
To the outside world, the acquisition looked like a giant company absorbing a smaller one. Internally, Raz describes it as a natural continuation of a long partnership.
Run:AI had worked with Nvidia for years. Their customers overlapped. Nvidia introduced Run:AI to enterprises who needed orchestration on top of Nvidia hardware. Customers often told Nvidia that Run:AI was the missing piece.
By the time discussions became serious, the relationship already existed. Nvidia, despite its scale, operated more like a 30,000-person startup. Decisions were fast, direct and founder-driven.
Raz emphasizes that Run:AI did not seek a buyer. Nvidia approached them.
Leaving a Successful Acquisition to Build Something New
You might expect Raz to stay at Nvidia after the acquisition. Instead, he decided to return to zero and build again. For years, he knew he wanted to start a company. He also knew who he wanted to build it with: a friend from his military service in the Prime Minister’s Office, where they worked on low-level cybersecurity systems.
They stayed connected for years, thinking differently from one another but collaborating well. When Raz felt it was time, the decision crystallized quickly.
“This is who I want to work with. This is my opportunity and this is the right timing,” he says.
Together they set out to answer a simple question:
Where can AI meaningfully change cybersecurity?
Why Identity Security Was the Obvious Target
Identity has been a security challenge for decades. Every company struggles with it. But Raz noticed something others overlooked: most identity decisions are still made by people guessing.
When someone requests access, a human must ask:
- Who is this person?
- What are they trying to do?
- Can they do it with fewer permissions?
- What do their peers have?
- What risks does this create?
The answers sit across scattered systems: HR tools, identity providers, documentation, permission charts. Humans rarely check all of these. They click approve and move on.
The result is friction between security and productivity, and an expanding attack surface.
“This is exactly where AI is strong,” Raz explains. “It pulls information from multiple sources, creates the context and makes smart recommendations.”
Fabrix is building a family of AI agents specialized for identity workflows such as access requests, lifecycle management, recertification and role design. They start by making recommendations. As trust builds, they automate more.
Integrating Without Forcing Behavioral Change
Raz learned early that organizations dislike changing their workflows. Even when excited about AI, customers say:
“Do it the way we work.”
Fabrix integrates into existing systems on day one. The AI learns from real activity before taking over decision making. Early customers in Israel helped prove that the system can deliver value without forcing teams to change their habits.
Why Identity Security Is Ripe for an AI-Native Approach
Identity tools exist, but Raz argues they were built for the pre-AI era. Large vendors would need to pivot their entire architecture to create true AI-native products. That is difficult for an established company with customers, liabilities and legacy software.
Startups have the advantage. They can design from scratch for a world where AI performs most of the investigative work.
He believes the real moat in AI is not lines of code. It is the organization itself.
“I am not building AI for security. I am building an organization that builds AI for security,” he explains.
AI-native companies prioritize model refinement, context engineering, evaluation pipelines and rapid iteration. Their R&D is structured around maximizing AI performance, not simply adding software features.
How Fabrix Stays Ahead in a Fast-Moving AI World
Raz and his team read constantly, track AI research and adopt new techniques early. Their engineers use AI tools to accelerate coding, prompt engineering and data processing. They test models on real customer data to measure depth and quality, not just synthetic benchmarks.
They also encourage aspiring AI builders to start quickly. Raz recommends no-code tools like n8n for building first agents, and advises newcomers to study both ML fundamentals and applied engineering.
The Israeli Advantage: Talent Density and Proximity
One theme appears throughout Raz’s story. Israel’s small size is a competitive edge.
Within a one-kilometer radius in Tel Aviv, thousands of companies operate, hire from each other, share knowledge and compete intensely. Raz compares it to a supergroup in music: many world-class performers concentrated in the same place.
This creates a feedback loop that strengthens new startups like Fabrix.
Where Fabrix Is Heading
Identity is becoming central to cybersecurity. Recent acquisitions, including Palo Alto Networks buying CyberArk, show how much the field matters.
Fabrix is starting with a narrow set of use cases, but Raz believes AI can eventually support every part of identity security. The company aims to expand quickly, grow its customer base and deliver quantifiable time and cost savings.
“Excited, definitely,” Raz says when asked how people should feel about the AI-powered future of identity.
Fabrix represents the next wave of Israeli cybersecurity: AI-native, fast-moving and founded by engineers who understand both the emotional and technical realities of building from zero.
If you’d like this adapted into a newsletter version, a social rollout or an SEO-oriented article, I can prepare those formats as well.
