How can we build AI systems we actually trust?
That was the question during a fireside chat at ETHSF’s “Verifiable Computation: From Zero-Knowledge to Infinite Intelligence” event. Moderator Kartik Talwar (A.Capital, ETH Global) sat down with Haseeb Qureshi (Dragonfly) and Daniel Marin (Nexus) to explore the real-world trajectory of verifiable AI and the cryptographic infrastructure underpinning it.
Their conversation ranged from the hard truths of deploying AI with crypto capabilities, to the strategic bets being made for what comes next. Below is a summary of the key ideas and some standout quotes from the conversation. The event was co-hosted by Nexus and Nethermind, sponsored by Halliday, and in partnership with Blockchain Builders Fund and ETHSF.
The state of crypto-AI: still mostly toys
While the convergence of crypto and AI has long been discussed, most real-world applications are still in their infancy.
“Everything starts as a toy. That was true for crypto, DeFi, AI — and it’s true for crypto-AI. These things look like proofs of concept until suddenly they don’t.”
— Haseeb Qureshi, Managing Partner, Dragonfly
The third wave: verifiability as a North Star
The panelists emphasized that the next meaningful evolution of crypto-AI will require more than entertainment.
Kartik Talwar sharpened the point: even with optimism about verifiability, the benchmark is clear — any decentralized alternative must eventually match or exceed what’s available from centralized incumbents like OpenAI.
“If it’s slower, more expensive, and worse quality — you’re just not going to use it. It needs to be better, or at least good enough.”
— Kartik Talwar, General Partner, A.Capital
The fourth wave: agentic AI that actually acts
What happens when AI agents can do more than just chat? The panel guests predict a coming phase where agents have wallets, take economic actions, and participate in onchain ecosystems — not just simulations.
“First it’s an animal in a cage. Then it’s shaking hands and doing business deals. And that’s when it gets interesting.”
— Haseeb Qureshi
Toward crypto-native benchmarks and business models
The panel closed by sketching some of the near-term infrastructure challenges and opportunities. From building crypto-specific AI benchmarks to developing agentic applications for sentiment tracking or market research, verifiability and specialization will matter.
And despite skepticism, the speakers agreed that crypto’s flair for financial entertainment — the gamified, memetic edge of the space — might play a critical role in bootstrapping usage.
“Someone is going to build the crypto-AI benchmark — a system that tests real agentic behavior in decentralized contexts. The only question is who and when.”
— Daniel Marin, Nexus
The takeaway
Verifiable AI may not yet be ready for prime time — but the clock is ticking. As centralized models grow more powerful and more opaque, the value of transparency will only grow. Builders and investors alike are betting that, when the crisis moment comes, the infrastructure will be ready.
“Don’t look away. Keep trying things. The biggest risk is not that it won’t work — it’s that you stop paying attention right before it does.”
— Haseeb Qureshi