The world of “Crypto + AI” can mean a thousand different things, but for simplicity, CBV divides the space into three buckets: AI-agent economy that’s powered by crypto; 2) Decentralized systems for training and building AI; and 3) AI-infused smart contracts.

I’ve been covering the Crypto + AI space since its inception (and hosted the inaugural AI Summit at Consensus), and I have to admit… this is the best organizing framework I’ve seen. So let’s quickly explore each bucket.

The idea is that all of us will someday be using AI agents for all kinds of tasks, our AI agents will need to spend money, and the most efficient way to do this is through cryptocurrency. “A significant amount of global GPD is going to be flowing through agentic systems,” says Coinbase Ventures’ Tejwani. “Crypto has very significant advantages to powering that. It’s permissionless. Any agent can spin it up.”

CBV’s initial investments in this vertical include Skyfire (which is building a financial stack for AI agents) and Payman (which is enabling agents to pay each other and humans). Payman’s co-founder, Tyllen Bicakcic, had the inspiration when his wife became pregnant nearly two years ago. “I started to think, what’s my daughter’s world going to look like when she’s 15 or 20 years old?” He suspected that in the future, AI Agents would need to actually pay human beings. Imagine that an AI agent could handle nine out of 10 tasks on a marketing mail campaign — brainstorming, writing, designing — but then needs a human to pick up the paper document and stuff it in a mailbox. (The name is almost cartoonishly explanatory. Payman.)

The idea is no longer purely theoretical. Bicakcic has built an early tool for developers, he says that 10,000 people are on the wait-list after his tweet announcing the project went viral, and he’s planning a demo this November in Bangkok, during Devcon, where an AI agent can buy you a beer. (I’ll be there to fact-check.)

“This second bucket is very vast,” says Tejwani in an almost comic understatement. This is the world of decentralized AI training, decentralized AI governance, and decentralized AI physical infrastructure (DePIN), which includes projects ranging from Bittensor to Gensyn to Aethir.

CBV is betting on decentralized data. Take one of their investments, a project called Vana, which incentivizes users to share data that can train AI. “Our core bet in the long run is that data is the mean differentiator,” says founder Anna Kazlauskas, who started mining Ethereum in 2015 and once kept a picture of Janet Yellen in her high school bedroom.

Here’s the working theory: In the arms race of AI, computing power will become table stakes. All the key players, inevitably, will pump more and more GPUs into the models, and it’s unlikely that anyone will have a clear edge. You spend billions to play the game, but that alone won’t win it.

You win the game with better data. But where does that training data come from? Much of the internet has already been hovered up by LLMs. Most of the books that have ever been written (at least the ones legally allowed to be used, and surely many that aren’t) are already part of the AI corpus. “Frontier models will hit a wall of data to train on,” says Tejwani. Models will be forced to rely more on synthetic data, meaning AI will create the data used to train AI — the dog eating its tail.

But there are pockets of data that Big Tech can’t touch, and blockchain could bring this data into play. Your medical data (for now) isn’t feeding any AI systems. Your financial data (for now) is untapped. Your LinkedIn posts (in theory) are not yet fed into the belly of OpenAI.

Through a system of tokens and DAOs, Vana lets you sell or rent your data — in a privacy-preserving way — to train AI. Parts of this puzzle are already up and running. Kazlauskas says Vana’s Reddit DAO (where users can essentially “sell” their Reddit comments to train AI) has 140,000 users, and that there are over 300 data DAOs on Vana’s test-net.

“Bucket 3 is where things get weird,” says Tejwani. “And we’re investing in the weird and the wonderful.” Tejwani envisions a world where AI is generating the vast majority of all software code. (We might be close already.) “Now apply that to these expressive smart contracts,” says Tejwani. “Almost in real-time, the AI is able to write smart contracts and build personalized on-chain apps.” This blends the real world, digital world, and blockchain world. Tejwani imagines that the AI could instantly design a front-end app with a silky-smooth UI, which would also “have the guarantee and power of smart contracts.”

The possibilities are dizzying, as is the very size of this new space. How many Crypto + AI projects are there? More seem to sprout from the ether every day, like memecoins. Tejwani guesses there are hundreds; Sokolin suspects there are between 250 and 1,000. I would guess over 300, partially because I have some unique visibility into this: Since I host the podcast AI-Curious and write for CoinDesk, I practically get a PR pitch almost daily (or at least weekly) from new Crypto + AI projects. They’re everywhere, and this isn’t always a good thing. “My guess is that 80% are theater, rather than anything substantive,” says Sokolin. “People enjoy taking the AI paintbrush and applying it to a token.”

On that note, it’s worth asking, how much of this is real? And does the world need any of this? Are projects just leaning into the AI hype machine? There’s a phrase you often hear in the Crypto + AI space: Crypto needs AI, and AI needs crypto. My hot take: This is only sort of true. For some broader perspective, it’s worth noting that when you speak to most innovators in the wider AI space, very few mention crypto or blockchain. (I’ve used the analogy before, but this reminds me of a one-sided college football rivalry; only one of the schools cares.)

Then again, look one level deeper, and the argument isn’t that AI needs blockchain to grow or scale or even thrive, but rather that if left unchecked, centralized AI will have extraordinary and perhaps even catastrophic influence. The very nature of truth is up for grabs. “In the future, if we get all of our information from AI, that would become the truth,” says Kazlauskas. “Whoever controls that AI model is controlling the truth.”

So who’s controlling that model?

On the current trajectory, the answer is a handful of tech monopolies. For the champions of Crypto + AI, the owners of the model look very different: Us. We would own the model. We would own AI.

Is this realistic, is this possible? Time will tell. But more and more, the biggest investors in the space are betting on these Davids against Goliath.