I dislike slide decks. Generally lots of fancy diagrams, bullet points, circles, arrows, garish colors and if we are lucky, a conclusion slide.
I used chatGPT to summarize the deck into something that consolidates all the information and ideas presented and makes more sense for me. Perhaps helpful for others also. I suggest that it would be good to include such a summary (or a button to generate such a summary) in future such decks. Thanks.
-----
From chatGPT:
"The deck’s core argument is that AI intelligence is becoming a commodity-like primitive, while applications are the layer that distributes that intelligence into the economy. As models get dramatically better and cheaper, the durable advantage for AI companies shifts away from simply having a better model and toward figuring out how to package that intelligence into products people will actually use and pay for. The interesting question is therefore less “who has the smartest model?” and more “what new products become possible as intelligence gets cheaper and more capable?”
For startups, this creates both a huge opportunity and a warning. Applications can capture value by embedding model capabilities deeply into specific workflows, industries, and consumer experiences—but generic AI features are likely to converge quickly as the underlying models improve. The winners will be products that turn continual model improvements into tangible customer value, with differentiated distribution, workflow integration, proprietary context/data, or a uniquely compelling user experience. In that sense, models provide the raw intelligence, while applications are the “diffusion layer” that determines where that intelligence actually gets deployed and monetized."
Like the materials in the deck ! But format is awful due to the colours, arrows and e.t.c (may be the structured plain text with bullet points would be better)
For health AI founders pitching in the current environment, this framing creates a positioning challenge. Investors who believe intelligence is the primitive increasingly ask why a specific health application cannot be replicated by a general model plus a good prompt. The answer requires demonstrating proprietary data, workflow integration, or outcome evidence that pure intelligence cannot substitute.
This is incredible work, and so happy to see this narrative being shared. I've spent the last eight months building an application-layer company on essentially this thesis, so this was a strange read in the best way.
The slide I keep coming back to is the verification one: "Reliable feedback is abundant in code and math; scarce in strategy, management and many real-world decisions." That line explains why coding got the first great agent companies, and it quietly predicts where the next ones get built — the domains where someone has to construct the feedback layer before diffusion can happen at all. Mine is brand marketing, which might be the extreme case.
If intelligence becomes a primitive, perhaps judgment becomes the moat. Applications can be copied; features converge. Knowing where to point intelligence may remain stubbornly scarce. Which is why I don’t recommend riding horses naked, seated backwards in a Western saddle.
I dislike slide decks. Generally lots of fancy diagrams, bullet points, circles, arrows, garish colors and if we are lucky, a conclusion slide.
I used chatGPT to summarize the deck into something that consolidates all the information and ideas presented and makes more sense for me. Perhaps helpful for others also. I suggest that it would be good to include such a summary (or a button to generate such a summary) in future such decks. Thanks.
-----
From chatGPT:
"The deck’s core argument is that AI intelligence is becoming a commodity-like primitive, while applications are the layer that distributes that intelligence into the economy. As models get dramatically better and cheaper, the durable advantage for AI companies shifts away from simply having a better model and toward figuring out how to package that intelligence into products people will actually use and pay for. The interesting question is therefore less “who has the smartest model?” and more “what new products become possible as intelligence gets cheaper and more capable?”
For startups, this creates both a huge opportunity and a warning. Applications can capture value by embedding model capabilities deeply into specific workflows, industries, and consumer experiences—but generic AI features are likely to converge quickly as the underlying models improve. The winners will be products that turn continual model improvements into tangible customer value, with differentiated distribution, workflow integration, proprietary context/data, or a uniquely compelling user experience. In that sense, models provide the raw intelligence, while applications are the “diffusion layer” that determines where that intelligence actually gets deployed and monetized."
Amazing read, minds brimming with ideas
Like the materials in the deck ! But format is awful due to the colours, arrows and e.t.c (may be the structured plain text with bullet points would be better)
Amazing read
For health AI founders pitching in the current environment, this framing creates a positioning challenge. Investors who believe intelligence is the primitive increasingly ask why a specific health application cannot be replicated by a general model plus a good prompt. The answer requires demonstrating proprietary data, workflow integration, or outcome evidence that pure intelligence cannot substitute.
This is incredible work, and so happy to see this narrative being shared. I've spent the last eight months building an application-layer company on essentially this thesis, so this was a strange read in the best way.
The slide I keep coming back to is the verification one: "Reliable feedback is abundant in code and math; scarce in strategy, management and many real-world decisions." That line explains why coding got the first great agent companies, and it quietly predicts where the next ones get built — the domains where someone has to construct the feedback layer before diffusion can happen at all. Mine is brand marketing, which might be the extreme case.
I wrote a full response from the operator — where the deck maps to what we see operating, and the follow-up question it leaves open: https://danrwilloughby.substack.com/p/the-diffusion-business-where-the
Really liked the format. Always looking to understand value shifts and barriers within the AI ecosystem. This was brilliant. Hope you keep publishing.
More texts on personal/consumer AI/agents!
India understands the mission https://www.facebook.com/watch/?v=2062439737947220
I really like this new format. Well done!
If intelligence becomes a primitive, perhaps judgment becomes the moat. Applications can be copied; features converge. Knowing where to point intelligence may remain stubbornly scarce. Which is why I don’t recommend riding horses naked, seated backwards in a Western saddle.