Happy Friday {{first_name}}!
Welcome to #77 of the Weekly AI edge:
The Big Story: Bittensor’s first halving finally hits.
The Alpha: Affine dethrones Chutes as the #1 Subnet.
The Weird: Beeple’s billionaire robot dogs.
Bittensor’s first halving lands on Sunday. I’ll dig into what this really means for investors later in the newsletter.
On a lighter note, I spent half an hour this week updating my Bittensor list on X, manually clicking through every new subnet and adding them. Feel free to follow the list to stay on top of Bittensor. I’ll keep it updated regularly.
The ecosystem is weird in a good way: niche agents, odd experiments, half-built products that shouldn’t exist but somehow do. The weirder the better - because it means something really cool could come out of it.
And speaking of weird… CT has been especially noisy lately.

Serpin clearly felt the same way and built something hilarious but genuinely useful: a tool that scraped 70k+ InfoFi users so you can get rid of the AI slop forever. It’s called shhhh, and honestly, the timing couldn’t be better.
The AI slop actually got loud enough that ZachXBT had to dig into its origins. About 26% traced back to Nigeria 👀
Last week’s poll results: 45.8% of you think Google takes the AI crown, with Chinese labs close behind at 37.5%. OpenAI barely cracked 8%. Didn’t expect that one.

🕹️Affine Dethrones Bittensor’s Top Subnet

Source: Taostats
A rare sight on the emissions chart this week: the top subnet just got dethroned… at least for a while.
Chutes has lived at #1 for as long as anyone has bothered to check the Bittensor subnet charts. If someone asked me “what’s the top subnet,” I didn’t think. I said Chutes. This week, Affine walked past it.
Affine focuses on reinforcement learning (RL) tasks, where models learn by trying things, failing, and getting better. It pays miners who make incremental improvements on a set of tasks (such as coding) and submit their models to Chutes.
Chutes is still ahead overall, but the gap is thinner than it has been in months.

🎭 Bittensor’s First Halving: What Actually Changes
The biggest thing happening in crypto x AI this month is Bittensor’s first halving. There’s a ton of noise flying around, so I want to clear the table.
On 14 December, TAO emissions will fall from 7,200 a day to 3,600. It’s the first real supply shift since the network came online. Like Bitcoin, TAO runs on a four-year halving schedule with a 21M cap.

Source: Grayscale
A lot of the confusion comes from people mixing up TAO emissions with miner rewards:
• Miner rewards don’t change. Miners earn subnet tokens, not TAO.
• Staking APR doesn’t change for the same reason.
• The halving only changes how much TAO flows into each subnet’s liquidity pool.
• The effect is slower TAO supply growth, not reduced payouts.
My take: 90% of CT expect a ‘halving pump’. In my opinion, these effects will play out over time, so I don’t expect much immediate volatility in TAO, but I do expect it to be bullish over the mid-term.
Fewer new TAO means less supply sell pressure (yay). Especially relevant when 70% of the supply is staked at 5%+ APY and subnet activity continues to grow.
Subnets will feel the pinch, though. With fewer TAO reaching each subnet’s liquidity pool, it will force subnets to become "businesses" focused on real utility, rewarding top performers more and filtering out low-value ones. This is good.
Institutional attention is also rising. Grayscale just published a full report on the halving, and a few funds have started taking structured subnet exposure. So the halving arrives right as outside interest is picking up.
Is the halving bullish or bearish for TAO?

DeAI led mindshare with a meaningful pickup. AI tokens mostly moved sideways ($20B market cap).
Robotics stayed firm on fundamentals: GEODNET’s ARR spike and peaq’s 6M+ users kept capital leaning toward hardware-linked names. Flows mostly concentrated in assets with either revenue or new launches.
🪙 Token Launches
🗓️ Upcoming Events
Gensyn (AI): Token sale begins on December 15
Kaito (KAITO): 1.18% token supply unlock on December 19

💸 Capital Flows
Hashgraph Ventures completed the first close of its $100M Web3 + AI fund LARA.
Surf raised $15M led by Pantera, with Coinbase Ventures and DCG to scale its crypto-focused AI models.
⚙️ Infra & Protocols
Brevis went live on mainnet with ProverNet, a decentralized ZK-proof processing network where users can earn by supplying proving compute.
Nous Research released Hermes 4.3, a 36B model matching Hermes 4 70B performance at half the size.
🤖 Agents & Apps in the Wild.
Inference Labs launched TruthTensor, a platform where AI agents trade on a simulated Polymarket portfolio.
NEAR and ADI launched TravAI, a travel agent that books full trips from a single natural-language request.
Fetch.ai launched Fetch Business, a platform for enterprises to create, deploy, and manage verified agents.
AITV launched its AI Streamer Factory on Base, letting users engage with AI streamers and earn on-chain rewards.
🧠 Bittensor Ecosystem
Chutes (SN64) TEE is now live, providing confidential compute capabilities to run and secure AI workloads.
SynthdataCo (SN50) upgraded its HFT prediction system, with 93% of miners adopting the new format.
Yuma announces its investment in Loosh (SN78), which is dedicated to developing emotional and cognitive intelligence for robots.
🦾 Robotics On-Chain
CodecFlow introduced MechaShip, a framework where delivery drones bid for jobs, prove delivery, and get paid trustlessly through ERC-8004 identities and Chainlink CRE.
Peaq launched a council and transparency page for its Network-Level Incentive Pools, opening pooled fee management to clearer oversight.
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 each week.

Xybot: The Robot That Works Onchain
Xyber has launched Xybot, a real physical robot that operates as an on-chain agent.
It executes tasks in the real world (inspections, monitoring, delivery) and settles the work on-chain with cryptographic verification.

Source: Xyber
The Wedge
How do you verify a physical robot actually did the job? Xybot’s control loop runs inside PROOF, a confidential compute layer that evaluates each step and logs it on-chain. You get a verifiable, tamper-proof record of the physical labor.
Its logic and actions are modular, almost like plugins. New “skills” can be added without rewriting firmware.
The Fine Print
On-chain verification solves the trust problem, not the physics problem. Sensors still drift, wheels get stuck in mud, and hardware breaks. The chain can prove the robot tried to execute the code, but it can't fix a broken motor.
Xybot represents the first step toward a Machine Economy, where robots work, earn and upgrade themselves. The idea is genuinely interesting, and I’d like to see more teams in web3 robotics explore this direction.

👋 This week’s deep dive is on INFINIT.
I’ve watched DeFi rebuild the same complexity moat that TradFi used to keep outsiders out.
What caught my attention about INFINIT, an AI agent protocol, is how it changes the equation: you express intent, and AI agents handle the reasoning and execution across protocols. I break down how that works, why coordination is the real bottleneck, and what it means for the future of agentic finance.
🔥 Our Weekly Top 5
#1 Beeple’s Billionaire Robot Dogs
Robot bodies, billionaire wax heads, and a machine that “poops” art. Creepy, funny, and a little too on-the-nose about who shapes culture.
#2 AI Traders Get Humbled by… Buy-and-Hold
Recall ran a live experiment to test a popular claim: “AI will replace traders.”
The result after 24 hours? Every model underperformed a simple buy-and-hold strategy.
#3 Kaito staking hits a reality check
Wale shared his Cysic airdrop payout after staking what was once $60k worth of Kaito. The reward came out to $96.
#4 The T800 had to prove it wasn’t CGI
So the EngineAI CEO let the robot kick him. A real hit, on camera. Probably the first time a founder has eaten a roundhouse to settle a Twitter argument.
#5 AI rewrites got smoked by human posts
Nine human-written posts ranked. Twelve AI-rewritten posts didn’t land a single one. Turns out readers filter out the ChatGPT sheen instantly.
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
If you enjoyed this, you’ll probably like the rest of what we do:
The Agent Angle: Weekly AI Agent newsletter & deep dives (non-crypto)
Our Decentralized AI canon 2025: our open library and industry reports
Prefer watching? Tune in on YouTube. You can also find me on X and LinkedIN
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.




