GM {{ First Name | }} 👋
It’s been a packed month over here, so I wanted to share a quick recap in case you missed anything.
We published 3 deep dives, kept the weekly AI Edge newsletter flowing (4x), launched a new newsletter focused entirely on AI agents, and kicked off a podcast!
Here’s what we’ve been up to:
1. World (WLD): Betting on Humanity
World was founded in 2020 by Sam Altman and Alex Blania. It doesn’t look like a typical crypto startup.
What they’ve built is more akin to a sovereign infrastructure stack with custom hardware (the Orb), biometric verification, a new L2 chain, and a privacy-preserving identity system, all to answer one question:
How do you prove someone’s a real, unique human at global scale, without sacrificing privacy?
Why we’re watching closely:
A real token sink is going live. Apps will start paying WLD for identity verification in Q3 2025.
The US launch unlocks faster onboarding and opens doors to partners like Visa.
Sentiment is deeply bearish. Every investor convo ends the same way: “Short every WLD pump.”
In our full thesis, we run through Bull, Base, and Bear scenarios for user growth and token burn by end-2026.
WLD still has brutal unlocks. But it’s a contrarian bet that verified digital identity will become foundational infrastructure.
Read the full piece: World (WLD), Betting on Humanity
2. Decentralized training is Crypto x AI’s moonshot.
There’s an emergence of a new player set in AI infrastructure: teams trying to train large models without centralized data centres.
With decentralized training, we’re redrawing the map of who can actually build foundational AI models.
To scale, these decentralized networks need to crack the “holy trinity”: Coordination, Verifiability & Performance over the open internet. No one has nailed all three yet.
Decentralized training blends parallelism strategies to manage memory, bandwidth, and efficiency. As models grow, model parallelism is resurging, as only the top-end GPUs can keep up with the memory requirements.
A focused group of teams is advancing the field, each with its own mix of compression tricks, and incentive design. Names to watch: Pluralis, Gensyn, Prime Intellect, Nous Research, Bagel, Templar, and Macrocosmos.
Read the full piece: Primer on Decentralized Training
3. State of the Swarm: The Language of Machines
As agent swarms evolve, their native language won’t be text. It’ll be code.
In the second piece of our series on AI agent swarms, we break down the four Acts we expect this shift to unfold through.
Most agent swarm systems today pose as open but are actually brittle, linear pipelines. Real flexibility demands more than plug-and-play. It calls for agents that can reason, adapt, and interoperate on the fly.
Read the full piece: State of the Swarm(II): The Language of Machines
We’ve launched The Agent Angle, a weekly newsletter on what’s breaking in AI agents. Each issue features five handpicked highlights: bold launches, weird demos, and research we think actually matters.
The debut edition is live.
Hint: The Eleven Labs agent we mentioned got many people talking. Read it, share it, and let us know what you think!
5. Escape Velocity Podcast
To wrap the month, I kicked off a podcast with my friend CK from Tensorplex Labs. We’re trying to make sense of AI, especially the web2 side of it: OpenAI, Gemini, AI doom discourse, and beyond.
More episodes (and some sharp guest interviews) are coming soon.
Have a great July ahead,
Teng Yan
