I’m Elias Martinez Cohen, and today I’m joining ORANGE JUICE as CTO.
I've worked in AI since 2014, back when it was mostly a research topic. A lot has changed over the past twelve years, but the last six months alone have flipped everything on its head. To many people, leaving NVIDIA for ORANGE JUICE may seem like a surprising move. Here's why I made it.
My background
I’ve spent my career building large, complex distributed systems that ship AI to the real world, managing petabytes of data and serving millions of requests per second.
In 2014, I was working at a startup that Twitter acquired, and we became Twitter Cortex. We went from a team of seven to a couple dozen engineers building the infrastructure that identified the contents of the thousands of photos people posted every second.
Most recently, I spent nearly eight years at NVIDIA building the data pipelines and workflow platforms behind its self-driving program, NDAS. If you drive a new Mercedes-Benz, its driver assistance was developed on my platforms.
You don’t need a big company anymore
Classifying photos in near real time at Twitter took years of work and enormous amounts of computing power. But today, the same basic task runs on your phone for free.
That arc is reshaping all software. It’s cheaper to build, easier to run, and increasingly personalized. Most small businesses have plenty of room for technological improvements, and now a small team can build bespoke tools for nearly every part of the business, from the front desk to the books, in days, at a price the business can afford.
Engineers with my background rarely work inside small and midsize businesses. A company with a hundred employees can’t economically justify a small army of engineers, so it buys ill-fitting off-the-shelf software and wrestles with it for years.
A holding company changes the equation, because one core technical team can serve an entire portfolio.
Back offices look a lot alike from one business to the next, even when the systems and details differ. What works in one company can work in others, and each win makes the next one cheaper. AI gives us an even longer lever. A handful of people can now deliver work that previously required an entire department.
Building impactful software used to mean working at a big company. That’s no longer true.
The adoption gap
The tools are here, but I see businesses struggling to use them. Recently, I spoke with a local business owner who runs a small design studio. When she heard what we’re doing at ORANGE JUICE, her eyes lit up and she started raving about her new friend, Claude. She’s in her 60s, has run the studio for 20 years, and has no plans to stop. Since the pandemic, though, good help has been hard to find. She isn’t running custom agent harnesses or orchestrating swarms or building her own applications. She has one Claude helping with email, meeting notes, and accounting, and it’s rejuvenated her business.
She’s only scratching the surface of what AI can do. I told her the next step is to completely automate away the admin work so she can spend all her time on design, where her passion and talent lie.
At NVIDIA, we worked closely with the frontier labs. I got early access to their newest models and tools, and I learned about bleeding-edge techniques from the engineers who built them. What is technically possible is far ahead of what most businesses use, and the gap keeps widening because capabilities improve faster than adoption. The field reinvents itself every 11 days. The typical owner can’t keep up with that while keeping the lights on.
No shot clock
Investors have noticed the gap too. Private equity firms and venture-backed roll-ups are buying small businesses and counting on AI-driven gains to justify the price. That’s a big bet at a time when AI is moving at a breakneck pace and few buyers know how they’ll actually put it to work.
ORANGE JUICE takes a more practical approach. We buy businesses that already have strong economics, so technology is upside on top of a sound model. That doesn’t mean AI is optional. Over the next decade, it will decide which businesses thrive, and my job is to ensure the companies we acquire are the ones that pull ahead.
As a permanent capital vehicle, we have no shot clock forcing an exit, which genuinely aligns everyone’s incentives. We’re building value that sticks, and compounding it over time into something bigger than the sum of its parts.
Why now
Leaving NVIDIA was one of the hardest professional decisions I’ve made. A lot of my work there was behind the scenes, building platforms for AI at scale.
At ORANGE JUICE, I get to help real people outside the tech industry navigate the AI revolution while it’s still early. Working at enterprise scale taught me a lot, but what excites me now is range: different businesses, many new problems, and software that takes the grind off the people running the show.
Right now, a small team has a unique advantage. So much can be automated that a few people can punch well above their weight. Big organizations carry a lot of inertia that’s hard to overcome, but a small team can be focused and move faster on bigger ambitions. We can spark a new idea on Monday and ship it on Friday the following Monday.
Best of all, I’m building ORANGE JUICE from the ground up with brilliant people I trust and truly enjoy working with.
Learn, build, repeat
Our model keeps sellers invested after the sale, so everything they know about running the business stays with it. We’ll change plenty over time, but we start by learning from the people who know the business best, not by arriving with a predetermined operating script.
A business is a lot like a distributed system: the individual parts depend on each other in ways that were never written down, and changing one piece often breaks another in unexpected ways. Before you change something, you must understand the dependencies and the invariants. You look for what’s hard, what’s slow, what’s unique to the domain, and where avoidable work creates friction. Then you fix what hurts most, one piece at a time.
We’re not here to automate away the people, or the parts of the job they love. We’re going after the drudgery. All software buys you time, and every business has hours we can buy back at a fraction of the cost.
Leading real world implementation
AI is the buzzword of the day, but it isn’t a panacea for business problems. Wherever possible, we will aim for a zero-token architecture: software that does the job without calling an LLM at all. Repeatable, clearly scoped work falls into this bucket. Traditional software is cheaper to run, more predictable, and easier to trust. AI now lets an experienced engineer write it much faster than before. Dynamic, unpredictable, and judgment-heavy processes may call for agentic systems that work alongside the team to increase operational efficiency.
There’s no universal playbook for any of this yet. We’re still very early, and AI has collapsed the cost of experimentation. We work alongside each company’s operators, test changes, measure results, and expand only what proves useful. That’s how we move fast without treating the core business as a laboratory.
For the founders and entrepreneurs who built these businesses, it should feel like reinforcements have arrived with state-of-the-art gear and the skills to build whatever the mission calls for.
If you own a business and you’re wondering how you’re going to steer your legacy through the AI transition, we’d love to hear from you. If you’re an engineer who wants to own real problems at businesses with real customers and decades of history, let’s talk. ORANGE JUICE isn’t going anywhere.
— Eli