Ori Goshen, Co-Founder and Co-CEO of AI21 Labs, on Dreaming Big and Changing the Future

Ori Goshen, Co-Founder and Co-CEO of AI21 Labs, has been at the forefront of technological innovation for over two decades, with a career spanning military-grade systems, groundbreaking startups and transformative AI advancements. In a wide-ranging discussion with Pitango, he shared his entrepreneurial journey, what’s needed to keep a tech company buzzing, his view on the future of AI, and how AI21’s vision was so ambitious they had to invent a market for it. Founded in 2017, AI21 Labs is redefining artificial intelligence by blending deep learning and symbolic reasoning to create smarter, more reliable systems. Spotting the tremendous potential in this forward-thinking approach, Pitango has proudly backed AI21 Labs from the innovative AI company’s earliest days. 

Let’s start with your background, Ori. What inspired you as a young tech enthusiast?

I grew up in Kfar Saba, Israel, the youngest in a family of electrical engineers. They were into hardware and I was the black sheep who loved software. My first computer – it was 1990 and I was four years old – opened up a world of possibilities for me.  Then, at 13, my brother brought me along to help out when he got a summer job fixing computers. I ended up being pretty good at ‘helping out,’ I guess, because the company hired me. It’s there that I was exposed to the early days of the internet. And by 16, I started my own boutique company building websites for other businesses.  My time in the IDF’s Intelligence Corps, though, was a real turning point. You arrive thinking you know so much and then suddenly you’re surrounded by these crazy smart guys, and you realize just how much more there is to learn. Moreover, the stakes were incredibly high – lives depended on the software we were creating.  It was there that I met Barak Lenz, who’s now our CTO at AI21 Labs. I remember one project we worked on together. People said it would take two years, something crazy like that – but we had no more than two weeks to deliver. We did it, somehow. Those kinds of intense challenges with a clear impact were really empowering and taught me about how innovation can thrive even under external constraints.  Overall, my time in the army shaped my understanding of personal responsibility while pushing my creativity to new limits.

After your military service, how did your journey into entrepreneurship develop?

I believe every domain has something interesting to offer, but for me passion is key. That’s why I joined Fring, a Pitango-backed startup, in 2007.  Fring flipped the script on mobile communication, with one of the first applications for free and accessible internet-based voice calls and messaging. On its first day on the market, we had over 100,000 installs and Fring quickly scaled to 40 million users.  That kind of immediate adoption and impact was thrilling, but watching how cool tech could fundamentally change people’s lives was a profound experience that has influenced everything I’ve done since. I went on to start my own company in 2009. We pivoted several times before moving into the analytics space, helping businesses make better use of their data, and eventually the company was acquired by another Israeli firm.

AI21 Labs has become a leader in the AI space. What motivated you to start the company and how did your vision take shape?

In 2017, my co-founders, Amnon Shashua and Yoav Shoham, and I took an unusual approach to starting a technology company. While most people begin by identifying a problem, we started with a vision: AI for the 21st century.  We noted that the incredible breakthrough of deep learning was great for pattern recognition, but life isn’t just patterns. AI-based systems need to be able to reason and understand context, as well. The integration of these two capabilities – we call it neurosymbolic AI – is at the heart of AI21. Statistical deep learning and rules-based logic are blended to create AI systems that are more reliable, relatable and capable of reasoning in ways that mimic human intelligence. We also decided to focus on language when most of the AI world was captivated by computer vision – such as for facial recognition or object detection.  We felt language was the true frontier, because it’s not just a tool for communication; it reflects how we think, reason and make decisions. If we could create AI that truly understands and generates language, then we could unlock entirely new capabilities.  Turning that vision into a reality depended on assembling the right team. Early on, we brought in Barak Lenz, who is an incredible technical leader. Yoav Shoham, who was a professor at Stanford, shared his experience and an instinct for where AI needed to go. We gathered some of the brightest minds in the field – people who were not only experts in deep learning, but who could also think beyond it.  So far beyond it, in fact, that at first there was no existing market for our technology – we had to invent one.  And so we did. Our first big step was WordTune, a writing assistant designed to help users improve their texts. It wasn’t the final destination, but it allowed us to refine our models and understand how people would interact with AI in the real world. The opportunity to pursue our larger ambitions came in 2020, when large language models (LLMs) were becoming more advanced. We decided to pivot toward the enterprise market and that’s where our focus has been ever since.

What unique impact will AI21 have on enterprises? 

Enterprises need AI systems they can trust to analyze complex information and integrate it into their workflows. We envision an “intelligence layer” – AI that synthesizes and shares knowledge seamlessly, enhancing collaboration and decision-making across large organizations. Imagine every meeting, for example, being recorded, analyzed and summarized by AI, instantly providing the relevant teams with actionable insights. That’s the kind of future we’re building toward.

When did Pitango enter the picture with AI21?

From day one Eyal Niv, Chemi Peres and Rami Kalish of Pitango were among the first to realize there was something big going on here. And Pitango has been a part of what we’ve been building ever since.  Pitango has shown a lot of trust in our team and fully supports our “dream big,” forward-looking mentality. Our relationship is more than an investment deal; it’s a deep partnership that AI21 can rely on no matter what entrepreneurial challenges come our way. It’s a relationship we are very fortunate to have. 

How do you ensure a stable culture of cutting-edge innovation in a fast-evolving industry like AI? 

The pace of evolution in our industry almost forces you to reinvent the company on a regular basis. As a friend of mine in the AI space commented recently, “I hope nothing happens this week that completely changes my roadmap.”  But one element in managing such a dynamic situation is to be brutally honest with everyone – employees, investors, the board – everyone. It’s all about the people, and being transparent about both opportunities and challenges is important to maintaining their trust and commitment.  And secondly, we’re always trying to nurture internal collaboration. When it is clear to everyone that we’re navigating this storm together, then motivation to cooperate increases.  I personally find it interesting to see how people’s perspectives evolve over time. They tend to become increasingly resilient in handling rapid changes, which contributes to our stability. 

The AI community has been buzzing about the comparative benefits of large language models (LLMs) and smaller, more efficient models. What’s your perspective on this trend?

It’s a fascinating time in AI because we’re seeing two seemingly contradictory trends at once. On the one hand, there’s the ongoing push to scale up LLMs – making them bigger, training them on more data, and increasing their sophistication. These large models have unlocked incredible capabilities and we’re still exploring the full extent of what they can do. On the other hand, we’re seeing a growing emphasis on smaller, more efficient models. They are not only cheaper to train, but they are also becoming just as capable, or in some cases even more capable, than the massive models of a few years ago. This is driven by advancements in optimization techniques, better training strategies, and the increasing use of specialized architectures.  As smaller models become more accessible, we’ll see a shift toward distributed AI and a more tailored, diverse ecosystem. I find the potential for decentralization particularly exciting. Ultimately, though, it’s about finding the right tool for the job. Both models have their place and the future of AI will likely involve a balance between the two. 

What’s your best advice for today’s entrepreneurs?

My biggest piece of advice is to avoid relying too heavily on playbooks. They’re tempting because they provide structure, but they’re often based on someone else’s journey in a different context. If you follow them blindly, you risk solving the wrong problems or creating new ones. The key is to understand the fundamentals of your business and then analyze from first principles – strip everything down to its basic truths and build up from there. If your approach doesn’t fit the playbook, that’s okay. Real innovation often happens off the beaten path.