GENESIS Pt. I: What The F*ck Happened to Crypto x AI?

Your favorite AI x crypto project is probably bullshit. Attention has evaporated. And maybe tokenized agents weren't bringing AGI to the blockchain. So where do we go from here?

Anyone that was around in the fall of 2024 on CT will remember two things; the run up leading to Trump's election and the AI meta. For a pretty sustained period, there were runners everywhere. A random dev dropping a coin for an agent with some hand-wavey credentials would see it run to $20M virtually instantly. People were getting hyperexcited about the Nth new agent launching that promised X, Y and Z, and they'd full port into it. Volumes on these coins got even large CEX's to list perps, and AI agent launchpads like Virtuals and ai16z pulled went 40-50x at peak.

Privates too were absolutely hot. Same pattern of founders with vaguely relevant credentials getting underwritten at very, very high valuations prior to shipping anything with very loose ideas just because they were sitting pretty at the intersection of crypto x AI.

It's not an exaggeration to say that the crypto x AI meta wave defined 2024 Q4. But half a year later, people barely speak about it. So what the f*ck happened?

An Enduring, Misguided Sector

I want to get one thing straight- obviously I am a very long-term believer that there is a massive opportunity at the intersection of crypto x AI, one that most people completely underestimate. If I wasn't, I wouldn't be working @ritualnet.

From my view a few things are true, though:

(1) There is a clear latent bid for AI exposure. With most of the hottest AI companies being private and inaccessible, and most public equity AI exposure rolling into big tech stocks or low-quality assets, one of the perennial promises of crypto x AI was the ability to create new protocols, assets, markets etc. that gave exposure to these markets. If that wasn't the case, the total agent market cap (per agent.cookies.fun) wouldn't still be sitting pretty at 10B+ even though most of them don't really do anything valuable.

(2) 95% of what exists in crypto x AI is vaporware.Yes, I know your favorite AI agent launchpad is worth over . Yes, I know you think that another network to incentivize better models-for-X will solve all your problems. But it's hard to ignore the facts. Darling projects (ai16z, zerebro, etc.) have been completely abandoned. Virtual's agent graduation rate is abysmally low, has seen a 97% drop in revenue, and now maybe makes $1-3k daily in fees. We've hit a critical point of saturation, where most agents that have been launched are functionally useless, don't have users and are just phantoms on the internet/annoying reply bots. Outside of that, big AI exposure coins (i.e. Grass, IO, etc.) look like this:

Even with something like Bittensor, which I generally consider to be a higher quality ecosystem that does have some well-meaning participants, sees an immense power law in the distribution of its subnets. There are subnets (i.e. Chutes) that seem to have real products/quality teams behind them and accordingly command a premium, but a majority of subnets (33 to be exact) sit below $5M mcap. Even with the higher range of subnets, there's a lack of clarity around expansion (i.e. GPU provisioning subnets offer AI compute at cents on the dollar comparatively, which are often subsidized via $TAO emissions. While this can and has been effective in user acquisition and network growth, I'll be watching to see how this converts to a sustainable long-term network). There's also many AI x crypto teams that raised at frothy valuations that I frankly think are DOA (dead on arrival), but I'm always in favor of giving people a shot.

(3) There *are* good teams (largely private) that are working on net-new things. While I am broadly skeptical around distributed training, I actually do hope teams like Prime Intellect, Nous, Pluralis etc. prove me wrong. They've demonstrated a fairly deep level of thought, a focus on core AI innovation and most importantly have invested a *lot* of time and resources into actually onboarding talent outside the space and getting them to work on hard problems. Teams like USDai, which are attempting to create synthetic stables backed by AI assets like GPUs, offer a pretty interesting vision and product for net new asset issuance based on underlying AI's assets. Some teams are attempting to tackle data collection for robotics, which is actually a hard task and there is a place in which incentive networks make a lot of sense over more traditional solutions/forms. I think most of the teams in this area are still privates.
Ultimately, I think the commonality amongst these teams is they (a) don't optimize for short-term games and (b) have worked hard or are working hard on creative a high density of talent pertinent to the sector of AI that they're working on.

Where Do We Go From Here?

Ultimately, the question any product needs to ask: is what we're doing actually creating some sort of meaningful value for users? Is this sustainable in the long run?

Sure, value could be entertainment/speculation and users could be consumers. It could also be an attempt at building a compute/data network to eventually platform and create value for enterprises. But it can be a lot, lot more.

I saw this tweet that talked about projects in the space that the author was tracking:

I thought it was pretty comprehensive so I categorized the projects into a few categories:

1. Autonomous financial agents (agent that trades for you, finds yield, etc.) --> most of the ones I could find had abysmal TVL, poor/nonexistent execution, or hadn't launched yet
2. Prediction and betting intelligence (agents that can bet on sports events or prediction market events) --> the one I did see that had a live PnL was negative, the rest are exploratory
3. AI-native financial infra (i.e. GPU backed stables) --> too early to tell if these will work
4. Resource aggregation (i.e. data/compute networks) --> effective at bootstrapping supply side, but most appear to be lacking demand side
5. Distributed Training --> early to see how this shapes out and if models will be competitive
6. Trading/Sentiment Intelligence/Trading --> tools for helping speculation
7. Robotics --> early to tell if these will work (some projects are working with big Robotics labs)
8. Consumer --> mostly chatbot rehashes with embedded speculatory dynamics

Mechanically, these can be condensed into three thematics: AI tooling/agents that promise to make you more money in markets, networks that dole out incentives to subsidize AI-critical resources, or tokenization of AI assets.

To be clear, I do think these are all very interesting categories, but there's a pretty glaring issue:

Almost all of these are done from the principle of "How can I fit AI into crypto product patterns?". In some cases, I think this can work (i.e. tokens can be very powerful for bootstrapping network effects for resource collection). In most cases, I think its the wrong question. The better question is "How do I build a product that can appeal to millions of users, today, that merge the best of crypto and AI?"

This divergence is why we've been spammed with endless tokenized agents, networks with supply/demand imbalances etc. They are just products that people just don't want to use. And it makes sense: crypto as an industry is known for taking the laziest path to execution. So what's missing?

We need teams and infrastructure that go beyond the meta. This means that instead of a "lazy" approach of "X works in crypto, so I'm going to solve Y in AI using it", we start thinking about products that are *only* possible with crypto and AI merged AND are net-new and differentiated from anything in crypto or AI.

It's pretty clear that the former approach has led to practically no users post a massive speculative frenzy, and you want products that benefit from, but are not reliant on speculative frenzy.

A lot of the work we've done at @ritualnet is towards platforming this future.

One core thesis of ours is that enabling users to directly access AI models from smart contracts opens up a design space for applications that are unique, in crypto or in AI. Ultimately, we want be the layer (alongside our ecosystem) on which users are actually owned, and not transient.

I'm excited to share a bunch of these soon but from a broader perspective, some shared patterns:

(1) "Lazy" tokenization isn't the end goal; if there is an agent, the end goal of it isn't necessarily tokenization, but for it to be a productive and core feature of an application (i.e. creation and resolution of flash prediction markets) that users actually find valuable.

(2) Focus on meeting users where they're at; while some of these platforms do appeal to speculators, many of them are focused on a set of users that exist even beyond crypto (the infrastructure for abstraction is certainly there to meet that demand). This also means being agnostic over infrastructure (i.e. Ritual allows access to closed and open-source models) to enable the best user experience.

(3) Going beyond crypto; we love CT and are active participants here, but the user base that can be captured for many of these applications is so much more expansive than our corner of the world. Many applications here are building for that + have deep team experience in sectors *outside* of crypto which inform their priors (i.e. a team building at the intersection of AI x manga, a rather large market) . We want to see applications that both CT and your everyday user loves.

(4) Primitive integration where it makes sense; things like financialization/incentives or privacy primitives shouldn't be forced on the user. Many of these applications use i.e. ZK or tokenization where it's important to the core product, but only where critical for users and platform growth.

There's a few more thematics, but this post is already getting long. You may have noticed I called this "GENESIS Pt. I".

This is the first of a series I'm doing on crypto x AI, starting with this post.

Next one will be diving deeper into the specifics of what I mean by actually net-new applications that need BOTH crypto and AI over a few different sectors, and how @ritualnet fits into this. I'm starting with an industry favorite: AI x speculation.

Stay tuned.