Decentralized AI Hype vs. Reality, w/ VCs Alex Odagiu & Daniel Barabander

What’s real in Crypto + AI—and what’s just noise? In this episode of THE PEOPLE'S AI podcast, presented by Vana, we sat down with two top investors to unpack the actual state of decentralized AI. These are the people who see all the decks, hear all the pitches, and are funding what they believe is the future of the space. So what’s legit, what’s frothy, and where’s it all headed?

First, we talk with Alex Odagiu, Investment Director at YZI Labs. Alex gives a clear-eyed view of the current landscape, sharing insights on the wave of projects flooding the space, the use cases that excite him (think data markets, DeFi agents, and composable infrastructure), and the challenges of signal vs. noise. He also breaks down how he uses AI tools in his own VC workflow—practical, actionable insights for anyone trying to level up their research game.

Then we shift to a wide-ranging, deeply thoughtful conversation with Daniel Barabander, GC and Investment Partner at Variant. We get into the big picture of how crypto and AI fit together, especially around agent composability, economic ownership, and verification layers. He walks us through why AI agents might need to spend cryptocurrency, where that thesis holds up, and where it doesn’t. We also explore what’s still broken in the space—and what’s needed for real breakthroughs.

Topics include:

[00:04:00] Why 75% of pitches Alex sees are Crypto + AI
[00:10:00] Why blockchain incentives are well-suited for labeling scarce data
[00:14:00] What’s broken in open-source model training—and how Web3 could fix it
[00:18:00] Agents in DeFi: useful or meme coin mania?
[00:21:00] How VCs vet AI x Crypto teams—and why founder ‘pivot mindset’ matters
[00:25:00] Practical ways VCs use LLMs for diligence and synthesis
[00:32:00] Daniel’s 3 pillars: Aggregation, Verification, and Self-Custody
[00:36:00] Do AI agents really need to spend cryptocurrency?
[00:44:00] Why economic ownership is crucial to decentralized AI
[00:48:00] Composability as a path to superintelligence
[00:55:00] What’s still broken: the “fuzzy verification” problem
[01:00:00] The underrated promise of modular data layers

We close with a fun look at the modular vs. monolithic debate—Daniel makes the case for why the open, decentralized, and composable internet still has a shot.

If you’re at Consensus and still in Toronto, swing by the AI Summit and say hello.

Daniel Barabander:

https://x.com/dbarabander

Alex Odagiu:

https://x.com/odagius

And a big thanks to our partner, Vana, whose mission is to enable user-owned AI through user-owned data. Vana recently announced a collaboration with Flower Labs to build the world’s first user-owned foundation model.

About Vana's collaboration with Flower:

https://www.vana.org/posts/vana-flower-labs-partnership

About Vana:

https://linktr.ee/vanahq

Vana on Twitter/X:

https://x.com/vana

Vana ecosystem: vana.com

Subscribe to The People's AI on YouTube:

https://www.youtube.com/channel/UCnLiYlJulQIcmvCjnVRYotw

Jeff Wilser on Twitter/X:

https://x.com/jeffwilser