luke

experience building @moonpay | 1% daily in AI, tokenization, & consumer

[luke @wavegod]

model subscriptions will be the loss leader for consumer AI

and the real business will be taking a cut of execution

[luke @wavegod]

The biggest consumer AI companies of the future won't charge consumers for AI at all. Change my mind.

Past two years: we have seen a huge push on frontier-level reasoning. Mathematics, research, coding, cybersecurity.

Next phase: we will see a massive push into the current most valuable area for AI, which is consumer.

"but chatGPT already has 1B users!! etc. etc."

Yes, it's true. But in order to chip away at profitability and create more valuable products for the world to use, companies need to make agents as seamless for everyday users as possible.

They absolutely cannot do that with expensive models tuned for frontier-level tasks; it's overkill on cost and usage. So what to do? Make smaller specialist models and you can pass savings on to the user.

"so how will labs and AI companies make money?"

The fundamental idea in our current market is that AI apps have a single stream of B2C revenue, and that is through subscriptions. I envision a future wherein agents like Muse, Dots, Duet, etc. are largely free, and any subscription is present to gate for super-users on running up egregious bills, where even then, they may be loss-leading.

"so how do they make money?"

The profit model will switch from subscription to execution and routing, where companies make money when their agents do things for people that involve value exchange, thus capitalizing on that value exchange by taking a cut.

Think about how much economic activity an agent could facilitate on behalf of a single person. Shopping, booking travel, ordering food, managing subscriptions, finding insurance, making payments, hiring services. The list goes on.

Every one of these actions represents a transaction that somebody is already making money on - the difference here is that instead of the consumer navigating dozens of apps and websites, their agent becomes the interface through which these transactions happen.

The agent isn't just saving the user time, but also directing demand.. and directing demand is an incredibly valuable position to be in.

This is already how much of the internet works.
- Google monetizes search intent
- Amazon monetizes commerce
- Payment networks monetize transactions
- Marketplaces take a cut of the activity they facilitate

Consumer agents have the potential to participate in all of these markets simultaneously.

There is an obvious question around incentives. If agents are compensated for routing transactions, how do we ensure they're acting in the consumer's best interest rather than whoever pays the highest commission?

I think this will be one of the defining challenges of the industry. Trust, transparency, and alignment with the consumer will be essential. An agent that consistently makes decisions against its user's interests will undoubtedly eventually lose that user. That is why I say that trust is part of the sell - whoever synthesizes and delivers trust as a service first skip ahead quickly.

But as these problems get solved, and the cost of intelligence continues to decline, the economics become increasingly compelling.

a consumer agent doesn't necessarily need to extract $20-200/month from its user to build a massive business. What it needs is to become useful enough that its user trusts it to handle meaningful economic activity and impactful decisions of all kinds.

The more useful it becomes, the more activity it facilitates. The more activity it facilitates, the more opportunities there are to monetize execution and direction of consciousness.

My prediction is that the most valuable consumer AI companies won't necessarily be those with the smartest underlying models - in fact, I doubt they'll even have models - probably just fantastic smart-routing.

They'll be the ones that own the consumer relationship, establish high levels of trust, and become the default interface between people and the rest of the economy.

The first phase of AI was about selling access to intelligence, and the next will be about monetizing what that intelligence actually does.

[luke @wavegod]

@beffjezos Needs better limits

[luke @wavegod]

@Henryf1w it is.

but the usage limits are rough

not because the limits themselves are bad, but because you have a similar limit to other subscriptions, spread across a multiple of processes

would not recommend in its current state for any heavy continuous technical work, even on Ultra

[luke @wavegod]

Everyone was against testing years ago, but now with AI, testing is one of the only productive ways to evaluate someone's understanding of something

[luke @wavegod]

@Muse Muse saved me high 4 figures in $ this week - thanks guys

[luke @wavegod]

@brandon_galang It's going to be really interesting to watch. As models commoditize and look increasingly like lower level infra (EC2 vs GCE, etc.), I think we'll see some large capture in the orchestration space

And we will see some amazing consumer apps

[luke @wavegod]

@Polymarket nice

[luke @wavegod]

I think models are verticalizing into specializations and years ago when the current naming conventions began, verticalization wasn't necessarily something that had a proven need - just general usability.

In actuality, it seems reasonable to believe that we will have many models under the hood of an AI product (think Muse or Dots), powering different parts that make the whole. i.e. Terra 6.1 for browser use, Opus 5.5 + Fable for advanced coding + pen testing (for those who even need it), maybe another one for real time voice, another for x, y, z.. and there will be rule-based orchestration taking place.

Right now, families and benchmarks are murky, but I believe the structure for the above will take form slowly in coming time.

Not to plug my own stuff but this post I just wrote provides some context re: where I'm coming from
https://t.co/f0BEfrWjEn

[luke @wavegod]

The biggest consumer AI companies of the future won't charge consumers for AI at all. Change my mind.

Past two years: we have seen a huge push on frontier-level reasoning. Mathematics, research, coding, cybersecurity.

Next phase: we will see a massive push into the current most valuable area for AI, which is consumer.

"but chatGPT already has 1B users!! etc. etc."

Yes, it's true. But in order to chip away at profitability and create more valuable products for the world to use, companies need to make agents as seamless for everyday users as possible.

They absolutely cannot do that with expensive models tuned for frontier-level tasks; it's overkill on cost and usage. So what to do? Make smaller specialist models and you can pass savings on to the user.

"so how will labs and AI companies make money?"

The fundamental idea in our current market is that AI apps have a single stream of B2C revenue, and that is through subscriptions. I envision a future wherein agents like Muse, Dots, Duet, etc. are largely free, and any subscription is present to gate for super-users on running up egregious bills, where even then, they may be loss-leading.

"so how do they make money?"

The profit model will switch from subscription to execution and routing, where companies make money when their agents do things for people that involve value exchange, thus capitalizing on that value exchange by taking a cut.

Think about how much economic activity an agent could facilitate on behalf of a single person. Shopping, booking travel, ordering food, managing subscriptions, finding insurance, making payments, hiring services. The list goes on.

Every one of these actions represents a transaction that somebody is already making money on - the difference here is that instead of the consumer navigating dozens of apps and websites, their agent becomes the interface through which these transactions happen.

The agent isn't just saving the user time, but also directing demand.. and directing demand is an incredibly valuable position to be in.

This is already how much of the internet works.
- Google monetizes search intent
- Amazon monetizes commerce
- Payment networks monetize transactions
- Marketplaces take a cut of the activity they facilitate

Consumer agents have the potential to participate in all of these markets simultaneously.

There is an obvious question around incentives. If agents are compensated for routing transactions, how do we ensure they're acting in the consumer's best interest rather than whoever pays the highest commission?

I think this will be one of the defining challenges of the industry. Trust, transparency, and alignment with the consumer will be essential. An agent that consistently makes decisions against its user's interests will undoubtedly eventually lose that user. That is why I say that trust is part of the sell - whoever synthesizes and delivers trust as a service first skip ahead quickly.

But as these problems get solved, and the cost of intelligence continues to decline, the economics become increasingly compelling.

a consumer agent doesn't necessarily need to extract $20-200/month from its user to build a massive business. What it needs is to become useful enough that its user trusts it to handle meaningful economic activity and impactful decisions of all kinds.

The more useful it becomes, the more activity it facilitates. The more activity it facilitates, the more opportunities there are to monetize execution and direction of consciousness.

My prediction is that the most valuable consumer AI companies won't necessarily be those with the smartest underlying models - in fact, I doubt they'll even have models - probably just fantastic smart-routing.

They'll be the ones that own the consumer relationship, establish high levels of trust, and become the default interface between people and the rest of the economy.

The first phase of AI was about selling access to intelligence, and the next will be about monetizing what that intelligence actually does.

[luke @wavegod]

@growing_daniel They aren't prevalent enough where I am

I know they're technically here, but I've seen like 5 ever

[luke @wavegod]

Lumara Film festival just paid out $32K to the Top 1 and you still think that AI video is doomed to fail?

[MoonPay 🟣 @moonpay]

the winner of the GRAND PRIZE at the LUMARA FILM FESTIVAL is...

CANDY by JIAZE LI

this out-of-this-world concept piece wins the award for best overall film, and comes with a $32,000 prize along with 3.2M Venice credits!

[luke @wavegod]

@EnvolDev better than I could do 🤷‍♂️

[luke @wavegod]

@jpshrodinger Terra is pretty much a directed Browser Use model

Does the same stuff at lower cost

implication is another step towards making it incredibly easy to deliver amazing consumer experiences for less

[luke @wavegod]

@MINHxDYNASTY the solution (probably)

[luke @wavegod]

@Jack_Raines Nice

[luke @wavegod]

@nvandewetering Terra is very underrated

[luke @wavegod]

@browser_use building for the consumer

[luke @wavegod]

@haider1 Terra is proof that the real consumer AI wave hasn't even hit yet

[luke @wavegod]

@ianlapham Saw a post today talking about how over the past X days, someone has built X*2 side projects and thought the exact same