Maor Farid, Founder and CEO of Leo AI – Interview Series

Dr. Maor Farid, founder and CEO of Leo AI, is an Israeli-US engineer, AI researcher, social activist, and entrepreneur. He conducted AI and mechanical engineering research at MIT as a Fulbright postdoctoral fellow and became the youngest PhD graduate in the history of the Technion – Israel Institute of Technology. He has built a community of 60K+ engineers and supports underserved youth through a nonprofit initiative.

Leo AI is the first AI for mechanical engineering – a large mechanical model for physical product design, enabling teams to turn ideas into production-ready 3D models in seconds. The platform helps companies reduce engineering time by 70% and accelerate time-to-market by 18%. Founded in 2023, Leo AI is already used by engineers at global companies, including Toyota, HP, Mobileye (by Intel), Philips, and Scania. Only a few months after the seed round (led by Flint Capital), its ARR grew by 300% in Q1.

You built your background across mechanical engineering, nonlinear dynamics, AI research, MIT, and the Technion before founding Leo AI. What led you to focus specifically on building AI for mechanical engineers, and what problem did you feel the industry was still failing to solve?

Honestly, frustration.

Before Leo, I worked as a mechanical engineer in defense, and I realized something wild: engineers spend an absurd amount of time doing everything but engineering. Seriously. We spend time perusing old folders, digging through vendor catalogs, looking for standards, manually reusing old designs, and asking a senior engineer who remembers why exactly this decision was made in 2011. You name it.

Software engineers have GitHub Copilot, and writers have ChatGPT. Meanwhile, mechanical engineers were still opening PDFs from shared drives called “FINAL_v7_REAL_FINAL.pdf”. The industry kept talking about “digital transformation,” but most engineering teams were still operating as if it were 1998.

That became the obsession behind Leo: can we build an AI that actually understands engineering? Not just language, but geometry, constraints, tolerances, manufacturing logic, physics. Something that engineers could trust with real work, not toy demos.

Because if you get a marketing paragraph wrong, nobody dies. But if you get a tolerance stack wrong in aerospace or medical devices, people absolutely can.

Why do general-purpose AI systems like ChatGPT and Gemini struggle with mechanical engineering tasks that require real-world physics, constraints, tolerances, and manufacturability?

Because they were never built for this, as generic AI models are trained mostly on internet-scale text: Reddit, blogs, Wikipedia, social media, and random forums. That’s fine if you’re writing emails or summarizing documents, but it’s a disaster if you’re calculating fatigue life on a welded bracket going into a defense system.

Mechanical engineering is not autocomplete. It’s constrained problem-solving under physics. A generic model cannot truly reason about manufacturability, thermal expansion, GD&T, material behavior, safety factors, or tolerance accumulation. Most of them cannot even open a CAD file natively.

The dangerous part is that despite all this, they sound convincing. Engineers are not anti-AI. They’re anti-BS. Right now, when it comes to engineering tasks, most generic AI systems are extremely fluent BS-generators.

That’s why we trained Leo AI differently, using over a million vetted engineering sources. We integrated it directly into engineering systems and made every answer traceable back to standards, formulas, and references that engineers can verify themselves.

Mechanical engineering has historically been slower to adopt AI than software development. What are the biggest barriers preventing engineers and manufacturers from fully embracing AI-driven workflows?

I think the biggest barrier is cultural trust. It is not technical at all. Software can fail and get patched tomorrow, but physical systems don’t work that way. If your AI-generated code crashes an app, users get annoyed. If your AI-generated engineering mistake ends up inside an aircraft, a medical implant, or a factory robot, the consequences are very different. 

Engineers are trained from day one to think about failure modes. We grow up hearing stories about bridges collapsing because somebody made the wrong assumption. So when Silicon Valley shows up, saying “just vibe engineer it,” mechanical engineers immediately reject it. 

The second barrier is that manufacturing companies are sitting on decades of undocumented tribal knowledge trapped inside PLMs, PDFs, CAD files, ERP systems, and retiring engineers’ heads. Generic AI cannot access or reason over that context.

And third: I don’t want to sound too brutal, but from my perspective, most AI products for industry are basically automation theater. Fancy dashboards on top of shallow models you’re unable actually to engineer with. Engineers see through that very quickly.

Leo AI focuses on what you call “Mechanical Intelligence.” What does that concept mean to you, and how does it differ from the broader wave of AI copilots entering the enterprise market?

“Mechanical intelligence” means AI that understands the physical world, not just language.

As I mentioned, most copilots today are fundamentally text systems. They summarize, rewrite, and generate content. This is useful, but still operates inside digital abstraction. Mechanical intelligence requires reasoning under physics, geometry, constraints, manufacturability, material behavior, assembly logic, cost, reliability, thermal performance, and safety. 

So, for us, mechanical intelligence means building systems that can responsibly participate in engineering workflows. It means reading CAD natively, understanding assemblies, solving equations, validating against standards, and connecting directly into PLM and ERP systems. 

How close are we to AI systems that can independently design highly complex machines such as jet engines, industrial robotics systems, or humanoids?

It is closer than most people think, though not quite the way Hollywood imagines it.

People picture a hero talking to a computer and a perfect machine appearing instantly. What’s actually happening is that AI is gradually removing repetitive layers in engineering, and it’s doing it so fast. So we get a well-designed projection linked to the proper documentation that humans can review and adjust – and with AI, this projection is done in minutes instead of months.

Could AI generate major portions of a jet architecture in the near future? Absolutely. We’ve tried this out on mock simulations with Leo AI, and we’re pretty close. But fully autonomous engineering without human oversight? I cannot foresee this happening anytime soon. AI will not replace engineers, but engineers using AI may replace those who do not.

AI infrastructure itself is creating major engineering challenges around energy consumption and thermal management. How do you see AI-driven mechanical engineering contributing to areas like advanced cooling systems and next-generation data center design?

One of the companies we work with, ZutaCore, builds waterless cooling systems for AI data centers, where thermal management is becoming one of the biggest bottlenecks to scale AI infrastructure. Their engineers faced a surprisingly expensive problem: every new deployment required manually redesigning pipe configurations to fit the system, which consumed engineering time and increased manufacturing complexity.

They asked Leo for a creative solution inspired by nature, and Leo helped generate a simple, adjustable pipe concept that eliminated the need to redesign the system for every project. Instead of custom manufacturing each time, the team could use standardized off-the-shelf parts. That reduced costs by roughly $400 per unit and eliminated an entire repetitive engineering phase from the workflow.

So, as we can see, AI is ready to solve some issues that were created by its own infrastructure. 

Engineering mistakes can have serious real-world consequences. How do you balance the speed and automation benefits of AI with the need for reliability, validation, and safety in engineering environments?

You never remove the engineer from accountability. Ever. That’s the core principle. We do not believe in “black box engineering”: every recommendation Leo gives is traceable, explainable, and verifiable. Engineers can inspect the source, formulas, standards, and assumptions.

In practice, the best AI systems in engineering are not replacing rigor. They are compressing the tedious work around rigor. The dangerous narrative right now is “speed at all costs.” That mentality works until you leave the digital world and start building physical systems. The physical world is unforgiving. 

You’ve said that AI will not replace engineers, but engineers using AI may replace those who do not. What new skills do you believe the next generation of mechanical engineers will need in order to remain competitive?

The most important skill will actually become deeper engineering judgment.

Ironically, as AI automates more of the execution work, human engineers become more responsible for defining constraints, validating outputs, understanding trade-offs, and catching failure modes.

Young engineers who blindly trust AI will become dangerous very quickly. The best engineers will be the ones who know how to orchestrate AI systems while still maintaining a deep understanding of first principles.

I think we’ll also see a huge shift toward systems thinking. Mechanical engineers will increasingly work, simultaneously, across software, electronics, manufacturing, simulation, and AI. The isolated mechanical engineer may disappear, but the multidisciplinary engineer will become extremely valuable.

We’re seeing increasing momentum around robotics, embodied AI, and physical AI systems. Do you believe the next major breakthrough in AI will come from systems interacting with the physical world rather than purely digital copilots?

Yes. I think we’re definitely moving this way, but we’re not quite there yet.

The first wave of AI was about language and information. The next one will be about interaction with reality. Once AI leaves the screen and enters the physical world, the difficulty level changes dramatically because reality introduces friction, uncertainty, material variability, and real safety consequences. The companies that win the next decade of AI will not just generate better. They will build systems that can reason about and interact with the physical world reliably.

As AI becomes more deeply integrated into engineering workflows, which parts of the design and innovation process do you believe will always require uniquely human creativity and judgment?

Responsibility. That’s the only answer. As I said, the physical world is unforgiving of engineering mistakes, and even at a very high level of AI reasoning, it will never be able to replace the human decision-making process.

AI can optimize, generate, research enormous design spaces faster than humans ever could — but humans should still decide what should exist in the world, what trade-offs are acceptable, what risks are ethical, and what constraints matter the most.

Honestly, some of the best engineering ideas come from intuition built over years of failure, experience, and weird human pattern recognition that’s very hard to formalize. So yes, I do not think AI will ever replace the human responsibility behind engineering decisions. That’s what makes it actually impossible to replace.

Thank you for the great interview, readers who wish to learn more should visit Leo AI.