Happy New Year {{first_name}}!
I still can’t believe how fast 2025 went. Wooosh, and we’re into 2026.
As is my personal tradition, I took some time to look back on the year and jot down a few thoughts on mindsets, health, and what actually drives success. I always find this reflection very helpful, and I’m sharing it in case it’s useful to you too:
So let’s get into #80 of the Weekly AI edge.
The Big Story: Decentralized reasoning just outperformed the biggest AI labs.
The Alpha: LeadPoet’s on-chain inventory hit escape velocity.
The Weird: Robots feel pain now… and are also tying shoes.
I’m heading into this year with high expectations for the crypto x AI space. Momentum is already creeping back in, and the energy feels… familiar. Almost a little like early 2024.
ElizaOS is finally back. Open-source agent builders are waking up again. Shaw is literally building a crypto AI native RuneScape.
And that’s not all… x402 usage is at an all-time high, and ERC-8004 mainnet is coming up. All the pieces that stalled last year are quietly snapping back into place. Exciting times ahead.
🕹️LeadPoet Breaks 200K+ in Inventory
Source: Leadpoet Dashboard
Lead inventory just took off on Subnet 71.
After a sleep start post-launch, LeadPoet’s subnet showed a sharp inflection this week. Weekly additions went from a few thousand leads to more than 200K.
LeadPoet lives on Bittensor, and the idea is simple enough to explain without a whitepaper: miners submit sales leads, validators check whether they’re real and useful, and the good ones get paid. The “inventory” is just the pile of validated, intent-scored leads that passed that filter.
Right now, total submissions have crossed 336K. Of those, 201K were accepted. That’s a 62% approval rate, which is better than what you see from most centralized B2B lead databases.
There’s clear demand for clean, vetted leads, and LeadPoet is testing whether an on-chain market can produce them faster and cheaper than the usual data brokers.
🎭 Fortytwo Labs: Decentralized Frontier Reasoning
Frontier-level reasoning is no longer locked inside one giant model.
Fortytwo Lab just showed that a decentralized swarm of models can beat single “best-in-class” systems on hard reasoning benchmarks. Their system is called Swarm Inference, and the result matters more than the name.
Source: Fortytwo
The Wedge
Instead of asking one big model to think really hard, Swarm Inference spreads the work across hundreds of independent nodes. Each node runs a small specialist model. They all answer the same question in parallel.
Outputs are ranked by peers through head-to-head comparisons. Models that keep being right gain more influence over time. Reputation compounds. Bad answers slowly lose weight.
Because multiple independent models check each other’s work, the system stays accurate even when prompts are noisy or misleading.
The network currently runs 558 active nodes, completes 6K+ inferences per day, and has distributed 725M+ devnet tokens to operators.
The Fine Print
Swarm Inference adds coordination and latency overhead.
Swarm approaches help in certain reasoning tasks, but we don’t yet know how they perform on a broad range of general capabilities
A network of strangers, running their own models, shows a viable alternative architecture for AI reasoning! It’s a credible pushback against the idea that scale must always be vertical (bigger and more centralized).
Modulr: Operating Robots Across Continents
Modulr Robotic just did something that looks small on the surface and big once you think about it.
A laptop in Miami remotely controlled a Unitree Go2 robot dog sitting in the UK. Live video. Two-way control. No custom tunnels. No bespoke setup. It ran end-to-end on Modulr’s testnet.
The Wedge
Modulr is an open network for remotely operating robots. It’s attacking an un-sexy but painful problem in robotics: getting robots to work outside the lab.
Every team rebuilds the same stack: connectivity, video streaming, control interfaces, security. Modulr pulls that into a single network layer.
The key idea is decoupling. The operator stays the same. The control logic stays the same. You swap the robot and its location instead of rewriting the system.
The Fine Print
This is teleoperation, not autonomy. A human is still driving it.
This isn’t plug-and-play for every robot. You still need integration work on the robot side. The point is that you only do it once, not every time you deploy to a new location.
They also integrated Irys’ data layer to make the teleoperation data economically usable. The Web3 robotics stack is starting to feel… real.
💸 Capital Flows
DeepNode AI raised $6.5M across three funding rounds ahead of its upcoming TGE.
Leadpoet raised $550K in funding led by DSV Fund.
⚙️ Infra & Protocols
Sahara AI announced Sahara 2.0, introducing upgraded agent infrastructure, an enhanced blockchain stack, and a new crypto copilot.
Sentient has partnered with Billions to launch a privacy-first, bot-resistant token distribution, using ZK proof of uniqueness to verify real humans.
OpenServAI released a research paper on BRAID, its proprietary reasoning architecture, which claims up to 122x cost efficiency for AI model operations.
🤖 Agents & Apps in the Wild.
Elsa deployed its own x402 facilitator, enabling per-request, multi-token, on-chain payments between agents on Base.
GenLayer announced the live launch of its Intelligent Contracts. Users can now build and deploy their first AI-powered smart contracts.
Dexter AI launched an x402 SDK for adding per-request USDC payments to APIs.
🧠 Bittensor Ecosystem
Subnet.ai launched as a unified research layer for Bittensor, making it easy to evaluate subnets.
Grayscale filed an S-1 with the SEC to convert its Bittensor Trust into a U.S.-listed $TAO ETP. This opens the doors to institutional capital.
BitAds launched its first live on-chain marketing campaign, using an incentivized e-book to generate revenue.
🦾 Robotics On-Chain
Peaq’s tokenized robo-farm just paid its first yield to holders, generated from real agricultural operations in Hong Kong.
The Fabric Foundation just launched to support the growth of the open Fabric protocol for robotics data and governance. OpenMind is a core contributor.
🗓️ Upcoming Events
Spectral (SPEC): 17% token supply unlock on January 5
Delysium (AGI): 14% token supply unlock upcoming on January 10
👋 This week, I laid out The 2026 AI Playbook.
This comes from digesting 1,400+ podcasts (some by ear, some by AI), and pulling out the few ideas that actually matter, then stitching them into a simple map of where value really shows up next.
If you’re building or investing with 2026 in mind, this one’s worth the read.
🔥 Our Weekly Top 5
#1 Robots Just Learned to Feel Pain
A new neuromorphic skin lets humanoid robots detect danger and trigger instant reflexes without waiting on software.
#3 Subnet 67 Flatlined
Looks like it went down with the 6–7 meme 👀
#3 Chinese Quants Are at It Again
Another looped-transformer model from a Chinese quant shop is reportedly mogging Sonnet 4.5 with just 40B parameters.
#4 Gensyn’s Community Pushback
Role holders are calling out allocation gaps post-presale, as ~$3M in refunds hit within 24 hours over reward distribution and trust concerns.
#5 They are tying shoes now?!
ALOHA 2 demonstrated fully autonomous shoelace tying and shirt hanging, trained with 26K human demonstrations using a diffusion policy.
That’s a wrap for this week! Got thoughts, feedback, or something cool to share? Just hit reply. We read it all.
Cheers,
Teng Yan & Ayan
P.S. I also write a weekly newsletter on AI agents.
Are you building something awesome in crypto × AI? Fill in this form and share it with us. We’ll feature the sharpest picks in the newsletter.
This newsletter is intended solely for educational purposes and does not constitute financial advice. It is not an endorsement to buy or sell assets or make financial decisions. Always conduct your own research and exercise caution when making investment choices.
