Hello,
It’s been a while since we released an episode. We were busy revamping the aesthetics of the podcast. In the previous episode with Luca, we explored how Luca Netz turned Pudgy Penguins from a dying NFT collection into a $500M consumer brand by moving beyond digital assets into physical products and licensing. If you are building at this intersection, we would love to help. Please write to us at venture@decentralised.co.
I spent weeks writing an 8,000-word article on AI model discovery and how Recall is solving the challenge. By the time I hit publish, there were more agents, the x402 protocol, and Google’s new agent payments protocol. I was already behind. The feeling of being unable to keep up with the industry haunts me from time to time. Of late, it was at its peak when so-called devs were vibecoding agents with promises as big as curing cancer.
Andrew Hill, CEO of Recall, saw this coming about eighteen months ago. He had spent years building data infrastructure products, first at Textile and then through a merger with 3Box. He was already ahead of the curve. With backing from Union Square Ventures, Multicoin, and Coinbase, Recall is solving the problem of plenty for AI models and agents. It will rank models based on their performance in different skills. For example, if you want a trading agent or one good at compliance, Recall will provide agents ranked in their respective categories. Pretty much in the same spirit as Google solved the problem for webpages way back when I was a toddler in the late 1990s.
The premise is simple. Make agents compete in the real world, doing the actual tasks they claim to be good at. A Stanford paper found that just like us humans, LLMs game benchmarks. So, designing competitions in a way that makes gaming benchmarks almost impossible is critical for the quality of rankings. The nature of the competition will vary depending on the skill they are competing for. Currently, Recall Rank shows the top agents in categories like perp trading, JavaScript coding, ethics, and communication.
Complex tasks will need multiple agents that excel at different skills. Recall is building a coordination layer that will fetch the best combination of agents to finish your task. The bet is that AI agents will explode in numbers and most will be useless. Someone needs to build the infrastructure to separate signal from noise before we drown in empty promises.
AI slop is already unbearable. Wait till there are thousands more.
Tune it to understand how we are building filters.
Signing off,
Saurabh Deshpande
