Yariv Adan spent 17 years building AI at Google, rising to Senior Director of Product Management. He co-founded Google Assistant and Google Lens, and led Google Cloud’s Conversational AI and Applied GenAI teams. After leaving in 2024, he founded Ellipsis Venture Capital, an early-stage fund that invests in what he calls “defensible AI” — companies with proprietary data or deep expertise that can withstand disruption in what he terms the era of “commoditized magic.” He was a speaker at a lunch in Davos, during the World Economic Forum’s annual meeting, that was organized by bluegain, a company that helps leaders with their digital business model and sustainability shifts. The Innovator caught up with Adan to discuss why AI is rewriting the basic rules of business.
Q: How, in your view, is AI transforming businesss?
YA: AI is rewriting the basic rules of what constitutes a good business, and the world is struggling to keep up. The entire economy today is based on this model of an industrial company that needs CAPEX and OPEX and raises it from the public markets. In the age of AI, you don’t need a lot of resources or that many people to run a very large company. You can automate a company almost at zero cost. What does that mean for the economy? For business? And for venture? Two years ago we coined the term ‘Commoditized Magic’ to describe this: the technology is truly magical — unlocking capabilities that were previously impossible — yet it’s almost entirely commoditized by frontier models. The magic creates massive opportunity. The commoditization creates massive risk.
Investors and entrepreneurs have been assuming that if you identify a real problem and you manage to build a team that can quickly build something of value by focusing on the user needs, it will give you a recipe for a successful startup. In the past, if you had a six-month head start, you had a first-mover advantage. But if that’s not the case anymore, what should I do as a founder? What should I do as an investor? The most basic rule of venture hasn’t changed: a company needs differentiation and defensible moats to sustain high-margin success at scale. But what counts as a defensible moat has shifted dramatically, with the bar rising to a much higher level. If your business lacks a genuine moat, whether proprietary data or unique expertise that can withstand an army of highly skilled AI agents, it will inevitably face disruption within the commoditization kill zone.
Q: Is this what we are now seeing the software sector, which is taking a beating on the stock market due to disruption fears?
YA: AI’s capability to generate high-quality code brings the marginal cost of creating software to near zero, as well as reducing the required time and expertise to build it. That changes everything. Consider: companies used to buy generic software, then spend years and millions customizing it. That complexity created lock-in. AI flips this entirely. We’re seeing commoditization rushing in from all sides simultaneously, and three forces are converging.
First, users themselves are becoming builders. A user can now go to ChatGPT, Claude, or Loveable and say ‘build me an app that does X.’ The software that previously required a license, integration consultants, and a three-year commitment can now be generated, used, and discarded. The cost of generating tens of thousands of lines of code is the same as the napkin they hand you at Starbucks, and you don’t save that napkin for next time either. I call these ‘ephemeral apps,’ and they represent an entirely new class of competition. A single user-builder is a formidable competitor to an entire SaaS company when it comes to building exactly the app they need at a given moment.
Second, as coding agents approach professional engineer-level capability at a fraction of the cost, the barrier to launching a software company drops dramatically. We see this in our deal flow every day. Every use case now has dozens of startups attacking it, each starting from a small beachhead of customers they already know, hoping to expand. But when they lift their heads, they see a hundred beachheads next to them, with no clear differentiation. These companies may deliver real value, some may even be profitable, but they don’t make sense as venture-backed businesses when crowding reaches this level.
And third — this is the part people underestimate — distribution moats are eroding too. The old playbook of ‘move fast, capture customers, lock them in with complex integrations’ stops working when AI can automate or regenerate those integrations on the fly. We’ll likely see procurement AI agents that bid and test for capabilities in real time, making brand and first-mover advantage largely irrelevant. When switching costs approach zero, loyalty follows.
On top of all this, Big Tech can also build applications much more quickly and with fewer resources, so they are moving up the stack. We see it with both the hyperscalers and the frontier models — they’re building vertical use cases that once belonged to startups, but starting from a billion users.
So yes, the stock market is reacting to something real: useful intelligence is becoming a commodity, and that changes everything about how software creates and captures value.
Q: If everything is moving to commodity, why did you become an early-stage investor in AI deeptech?
YA: There is clearly a massive opportunity for new unicorns — just with a higher bar for defensibility. If you look at previous tech revolutions, the biggest winners weren’t doing old things slightly more efficiently; they created entirely new businesses. That’s what we’re looking for. There are two main problem spaces we pursue:
First: high-value problems that were previously considered impossible are now unlocked with recent AI capabilities — creating a strong “why now.” Second is what we call disruptive AI-first technology — small teams of top world experts building critical parts of the emerging new AI stack that Big Tech won’t build, but that are critical to unlocking the value of AI at scale.
For both of these we require clear defensibility against the commoditization forces I mentioned. We believe the only real moat is data — problems that require unique proprietary data that AI models don’t have and can’t easily scrape or synthesize, and where there is a clear data flywheel that increases value and stickiness over time. We love companies that combine AI deep tech with deep industry or science expertise.
For example, we’ve invested in a company automating and optimizing manufacturing in plants — reducing energy, increasing throughput, and over time optimizing the entire plant. The datasets required for training are proprietary, and deploying the product generates new proprietary data that makes the model better. That’s a real flywheel. We invested in another company that uses AI to design unique and very complex proteins that can be used to unlock new immune therapies that transform care for cancer, autoimmune disease, and other severe conditions. Even with recent breakthroughs in protein design, only a very small subset of all potential proteins has been explored, and doing this accurately and efficiently is still a very hard problem, requiring unique data and expertise.
Q: You recently wrote an article entitled “Donkeys, Not Unicorns.” What did you mean?
YA: The superpowers of AI have unlocked an avenue for entrepreneurs that doesn’t require venture capital at all. What if, instead of chasing one elusive unicorn, you used AI agents to automate and scale a portfolio of value-generating businesses? Leverage their product and go-to-market focus on an underserved niche or market segment, use the efficiency of AI to do it at minimal cost, time, and resources — maybe even run multiple such bootstrapped, AI-native companies.
Imagine a small team running multiple micro-businesses simultaneously — one automating compliance reports for European fintechs, another generating training materials for logistics companies, a third managing invoicing for freelance consultants. None is a billion-dollar market. None lands a TechCrunch headline. But each generates steady revenue, and together they compound. The founder isn’t managing fifteen teams; AI agents handle the build, iteration, and support. The founder’s job is portfolio management: which donkeys to feed, which to retire.
Q: AI is, of course, impacting traditional businesses as well. What advice do you have for them?
YA: Manufacturing companies respond to my analysis by saying, ‘You are talking about software companies, but I don’t have a problem because I build cars, or I build this or that.’ What they are ignoring is that AI automation and efficiency is impacting both the bottom line and the top line in a very big way. Even if you are building a physical product, if you are not using AI in your back office to cut costs, or as part of your manufacturing process to save energy or increase throughput, or in your business outreach, you will fall behind other companies in your sector that are using AI very effectively. If you’re not investing now to be able to do this in two to three years, even if your core product is defensible, you won’t be competitive.
To become an AI-first company you need new talent, new processes, new data, and new development practices. I think in a best-case scenario this takes two to three years. If you take a wait-and-see approach like Yahoo or Nokia, it takes a decade and by then you are irrelevant. Time to adapt is a reality, but I think the biggest challenge is talent because the talent doesn’t want to work in these traditional companies. My very strong recommendation is: don’t worry about getting value from AI in the short term. Getting value from AI is tactical, but making sure that you are still relevant in three to five years — that’s strategic. So start by understanding how AI can help your business and develop a vision so you can start attracting the talent you need. That should be the top priority. I think large established corporates do have an advantage in the short term because the startups have a distribution challenge: they need to find their customers and still need to learn the market. Large corporates are embedded in the market. They have the customers and they have data and can leverage these assets.
My advice is to start developing your AI strategy, invest, and get a foot in the door to ensure that you are a relevant competitor in the next phase. Don’t hold on to your old wins and your old way of doing things. Take a step back and start thinking about whether you will have a sustainable, defensible moat once the startups gunning for your space solve the problem of distribution. Many companies overestimate their moat. Start thinking about pivoting towards new areas where you can leverage your proprietary data and proprietary expertise. In sum, all traditional companies need to invest in AI. Don’t worry about value in the short term. Rather than doing a bunch of proof-of-concept trials, think strategically and get very serious about preparing for the future.
This article is content that would normally only be available to subscribers. Become a subscriber to see what you have been missing
