Great piece Andrew. One interesting angle: What if the real moat isn't just adding network effects to AI, but AI's ability to dramatically lower the activation energy for network effects? Traditional products needed significant user behavior change to create networks (posting, sharing, connecting). But AI could automate much of this friction away - imagine AI auto-generating shareable content or proactively connecting relevant users. The next Facebook might not need to convince users to build their networks manually.
great point -- yes, and you could argue that machine learning's already been used for many relevant use cases, such as "People you may know" and feed algos for retention, etc. We already know this is effective.
The main issue for startups is that early on, when they don't have many users, these optimizations may not help **that** much since they just need to acquire more people and get product market fit. So it may end up feeling more of an optimization for new entrants, whereas it might help for the incumbent players
One wrinkle worth thinking about: unlike Web 1.0 and 2.0, where incumbents were slow to react, today’s corporate leaders are digital natives who fully grasp the power of AI. They aren’t just sitting on their hands—they’re aggressively investing, locking down proprietary data, and leveraging their distribution advantage.
This game won’t be as open as past cycles. The real question is: if incumbents hoard data and distribution, what’s the new edge for startups? Will it come from novel UX, emergent behaviors, or entirely new business models? Feels like a harder chessboard than before.
I generally agree that the incumbents have a leg up in the AI phase in a way that they didn't in mobile. Mobile was also pretty quickly adopted by incumbents too, it's just that it requires a full rewrite of the UX to adapt to the new device profile, whereas AI can more likely be added on.
That said, its been surprising how fragmented the large model layer has been, how many folks are going the API and open source route, etc. If you want to start a new AI-first app company, there's a ton of choices for which model to build upon.
The new UX innovation is an interesting question though. Might be that conversational is very important, and there's something new there. Or maybe some second-derivative format like avatars or something. We'll see. Tech will march ahead, so I have to bet there are new modalities in each generation, not just more of the same.
Brand = A Emotional + Logical Choice In Chaos. Brand is the Beacon in darkness. If this kind of speak triggers (then you don't under stand Brand as a value proposition in itself) or miss what Andrew is saying here. We, the customer (human mode not tech mode) walk into a store and see 100 pairs of Blue Jeans and we have to make a choice. How do we do so? We make an Heart + Mind choice. We look for indicators. Now granted Levis was first mover but 150 years later the 'defense–ability' of Levis is mostly Brand (and the occasional innovation.) If they don't spend and build brand that 150 years of blood sweat and tears and money goes away with some new fad but somehow we are still talking Levis today — thanks to the Brand aka its Wrapper Tactics.
Now can innovation in itself be a brand? Yes but thats for special ones like Apple was or Nike is sorta is. If we don't see [brand] then we the customer look for commodity pricing and commodity is all about cheap. For Brands we will happily pay for the AMG package — for cannibalizing commodities we will happily shop the sale. This is no different with AI. Get the wrapper right Ladies + Gents.
🌩Who ever owns Brand aka hearts + minds will gain Trust. And Trust in this emerging market is the Moat. In emerging arenas like AI, where innovation abounds yet differentiation is rare, getting the wrapper right is everything. Own the hearts and minds of your customers, and you build an impenetrable moat of trust that competitors can only dream of matching and yeah like Andrew said tap into the Network that have these [primed] Effects for increased results. 💡
Andrew we agree with you brother🧮. Important read you sent out with this article — got the coffee going already here in NYC.
Assuming the system layer remains the most competitive space in a generation, if maybe ever, we will stay focused on differentiation that can be built at the data/app layers as well as (I think actually most importantly), "AI Native Founders" who are building more efficiently than ever before with these tools at their side.
Do the LLMs become more like the microprocessor in the background or do they become the actual user product? If the models get commoditized, customer love and switching costs (including community) for specialized experiences will become the durable moats.
Which multiplies the # of viable ventures one can imagine.
Thanks for the insightful write up, Andrew! The described situation fits well the overall trend of diminishing costs in the "virtual world".
It's already well-established fact that costs of any virtual resource are diminishing. Some of these resources may be bound to physical resources, such as cloud compute or storage; other resources have value only because of agreed rules of artificial scarcity - from bitcoins to starships in Eve to paywalled news articles. Yet tendency for both is the same.
AI merely enables the same diminishing costs for "virtual labor". If we think about it, it was somewhat weird that some products had barrier to replicate merely because of "virtual labor" such as software engineering.
Advancement of AI puts the "virtual world" to where it intuitively belongs - costs of creating anything virtual should be diminishing, just because the thing is virtual, imaginary, not physical.
Innovation and emergence of new products will still take place, just the focus will move back to physical world, which is hard, messy, requires actual effort to move stuff, to store stuff, to protect against theft and fire, etc.. It still will be hard to build next Uber or DoorDash, as you need to mobilize real cars, couriers, people or robots, carry around real food that gets cold or stale, etc..
Great piece Andrew. One interesting angle: What if the real moat isn't just adding network effects to AI, but AI's ability to dramatically lower the activation energy for network effects? Traditional products needed significant user behavior change to create networks (posting, sharing, connecting). But AI could automate much of this friction away - imagine AI auto-generating shareable content or proactively connecting relevant users. The next Facebook might not need to convince users to build their networks manually.
great point -- yes, and you could argue that machine learning's already been used for many relevant use cases, such as "People you may know" and feed algos for retention, etc. We already know this is effective.
The main issue for startups is that early on, when they don't have many users, these optimizations may not help **that** much since they just need to acquire more people and get product market fit. So it may end up feeling more of an optimization for new entrants, whereas it might help for the incumbent players
One wrinkle worth thinking about: unlike Web 1.0 and 2.0, where incumbents were slow to react, today’s corporate leaders are digital natives who fully grasp the power of AI. They aren’t just sitting on their hands—they’re aggressively investing, locking down proprietary data, and leveraging their distribution advantage.
This game won’t be as open as past cycles. The real question is: if incumbents hoard data and distribution, what’s the new edge for startups? Will it come from novel UX, emergent behaviors, or entirely new business models? Feels like a harder chessboard than before.
I generally agree that the incumbents have a leg up in the AI phase in a way that they didn't in mobile. Mobile was also pretty quickly adopted by incumbents too, it's just that it requires a full rewrite of the UX to adapt to the new device profile, whereas AI can more likely be added on.
That said, its been surprising how fragmented the large model layer has been, how many folks are going the API and open source route, etc. If you want to start a new AI-first app company, there's a ton of choices for which model to build upon.
The new UX innovation is an interesting question though. Might be that conversational is very important, and there's something new there. Or maybe some second-derivative format like avatars or something. We'll see. Tech will march ahead, so I have to bet there are new modalities in each generation, not just more of the same.
🛸 Brand is the only Moat.
Brand = A Emotional + Logical Choice In Chaos. Brand is the Beacon in darkness. If this kind of speak triggers (then you don't under stand Brand as a value proposition in itself) or miss what Andrew is saying here. We, the customer (human mode not tech mode) walk into a store and see 100 pairs of Blue Jeans and we have to make a choice. How do we do so? We make an Heart + Mind choice. We look for indicators. Now granted Levis was first mover but 150 years later the 'defense–ability' of Levis is mostly Brand (and the occasional innovation.) If they don't spend and build brand that 150 years of blood sweat and tears and money goes away with some new fad but somehow we are still talking Levis today — thanks to the Brand aka its Wrapper Tactics.
Now can innovation in itself be a brand? Yes but thats for special ones like Apple was or Nike is sorta is. If we don't see [brand] then we the customer look for commodity pricing and commodity is all about cheap. For Brands we will happily pay for the AMG package — for cannibalizing commodities we will happily shop the sale. This is no different with AI. Get the wrapper right Ladies + Gents.
🌩Who ever owns Brand aka hearts + minds will gain Trust. And Trust in this emerging market is the Moat. In emerging arenas like AI, where innovation abounds yet differentiation is rare, getting the wrapper right is everything. Own the hearts and minds of your customers, and you build an impenetrable moat of trust that competitors can only dream of matching and yeah like Andrew said tap into the Network that have these [primed] Effects for increased results. 💡
Andrew we agree with you brother🧮. Important read you sent out with this article — got the coffee going already here in NYC.
Best read all week Andrew -- we got furrowed brows when we asked folks not to hate on GPT wrappers earlier last year -> https://threerockers.substack.com/p/dont-hate-on-gpt-wrappers.
Assuming the system layer remains the most competitive space in a generation, if maybe ever, we will stay focused on differentiation that can be built at the data/app layers as well as (I think actually most importantly), "AI Native Founders" who are building more efficiently than ever before with these tools at their side.
Do the LLMs become more like the microprocessor in the background or do they become the actual user product? If the models get commoditized, customer love and switching costs (including community) for specialized experiences will become the durable moats.
Which multiplies the # of viable ventures one can imagine.
Thanks for the insightful write up, Andrew! The described situation fits well the overall trend of diminishing costs in the "virtual world".
It's already well-established fact that costs of any virtual resource are diminishing. Some of these resources may be bound to physical resources, such as cloud compute or storage; other resources have value only because of agreed rules of artificial scarcity - from bitcoins to starships in Eve to paywalled news articles. Yet tendency for both is the same.
AI merely enables the same diminishing costs for "virtual labor". If we think about it, it was somewhat weird that some products had barrier to replicate merely because of "virtual labor" such as software engineering.
Advancement of AI puts the "virtual world" to where it intuitively belongs - costs of creating anything virtual should be diminishing, just because the thing is virtual, imaginary, not physical.
Innovation and emergence of new products will still take place, just the focus will move back to physical world, which is hard, messy, requires actual effort to move stuff, to store stuff, to protect against theft and fire, etc.. It still will be hard to build next Uber or DoorDash, as you need to mobilize real cars, couriers, people or robots, carry around real food that gets cold or stale, etc..
Very insightful and inspiring piece! Literally couldn’t stop reading, thank you!