Happy Thanksgiving weekend {{first_name}}!
So I just want to say that I’m grateful you make time for this every week. Your attention is not cheap, and I never take it for granted. 🫶
Welcome to #75 of the AI Edge.
I learned this week that a single telegram notification while you’re writing code or reading papers steals the next 23+ mins of your best thinking. Check your phone 40x a day, and you’ve burned 15 hours a week of peak performance.
I read that and turned off my notifications while working (‘do not disturb’ or ‘focus’ mode is the easiest way). Instantly felt better. You might want to test it too.
Focus turns out to be the week's main theme. Washington revealed Project Genesis, the closest thing to a Manhattan Project since 1939. It’s a push to expand America’s lead in frontier AI by treating compute and model training as strategic priorities. The timing makes sense given how fast China has been pushing new systems.
When governments start naming things Genesis, the memes basically write themselves
Also: Polymarket now thinks the AI bubble isn’t popping anytime soon. Yes, the bubble has found its way into the prediction markets. I’m taking a poll of our readers - what do you all think?
Are we in a huge AI bubble?
Let’s dive in.
It’s been one of those weeks where the market cooled off a bit (finally). AI slipped to a yearly low in CT mindshare. AI tokens pulled back 2% to $21.5B and robotics fell 9% to $1.05B.
Mindshare rotated toward ZK and EVM as Monad mainnet arrived. General AI usage hit new highs with Gemini 3’s launch and Grok crossing 10M daily visits.
$KITE +21% post-Binance listing recovery fueled by AI-agent economy hype and fresh liquidity inflows. I previously wrote a deep dive on Kite AI here.
$ONO +75% breakout from consolidation on surging DePIN network revenue and miner accumulation.
🪙 Token Launches
🗓️ Upcoming Events
Almanak: Strategy Builder v1 launches on November 29
🧪 Gensyn Testnet Transactions Cross 500K
Source: Gensyn Testnet Explorer
The kind of chart that makes you pause and look twice…
Gensyn’s public testnet just pushed past 500K daily transactions, climbing from ~350K in under a week. Total activity is now above 84M+ transactions with more than 155K users.
Most of this traffic comes from its early ML apps:
RL Swarm tests distributed reinforcement learning (33K+ connected nodes)
BlockAssist captures behavior from interactive environments (1M+ trained models)
CodeAssist captures edits and reasoning steps from coding tasks. (55K+ solved challenges)
Most of today’s usage still comes from Gensyn’s own applications, so the real test will be how much external developer activity shows up once the network fully opens.
The testnet did hit an RPC issue earlier this week that caused some nodes to disconnect and drop, but the chain itself kept running, and the fix only required a restart.
📟 CodecFlow: Operators Instead of Scripts
A few months ago, I came across CodecFlow, an execution layer for AI agents and robots. Their approach centers on an ‘Optr’ (or Operator) model that lets you apply the same automation logic to a desktop screen, or a robot arm without rewriting code.
Source: Codec Docs
The Wedge
An Operator is a small AI that watches what’s happening on a screen or in a robot’s camera, thinks about what it sees, and then takes the next step. It repeats that loop over and over.
Each Operator runs on its own little computer, streaming video to the AI and sending back actions such as clicks, keystrokes, or robot movements.
Teaching an Operator is basically showing it how a task should be done. You walk through it once or twice, and the system learns from your examples. When it’s ready, you can publish it so someone else can run the same Operator.
Source: Codec Docs
The Fine Print
Operators rely on consistent behavior across different machines. Real robots and real environments don’t always behave the same way, so unexpected edge cases will show up quickly.
Codec also built Fabric, which spins up the machines Operators run on and moves them between cloud, local computers, and real robots without interrupting the task. The real draw is that every agent runs in its own secure, isolated environment, so it can do the job cleanly without noise.
Project Genesis = the Manhattan Project of the 21st century?
Are you building something awesome in crypto × AI? Or spotted a startup or product that more people should know about? Fill in this form and share it with us. We’ll feature the sharpest picks in the newsletter each week.
💸 Capital Flows
Numerai raised a $30M Series C at a $500M valuation to scale its AI-driven hedge fund.
AlphaTON Capital secured an $82.5M investment to build out GPU infrastructure for its Cocoon AI project (related to TON).
ZENi raised a $1.5M Seed round to build its intelligence data layer for AI agents.
⚙️ Infra & Protocols
Prime Intellect just dropped INTELLECT-3, a fully open-source 100B+ MoE RL model that delivers SOTA performance for its size across math, code, and reasoning.
Monad launches AI blueprint, a full-support program that gives AI apps the resources and infrastructure to build, launch, and scale on Monad. Applications are now open.
Akash Network launched AkashML, a managed inference service for running open-source models on its decentralized GPU market.
Caesar partnered with Centrifuge to begin issuing on-chain equity, becoming the first AI company to take this step.
AO’s mainnet is now fully operational, completing the migration of legacy compute to its Hyperbeam architecture.
Fleek launched its Weyl Gen AI API, offering diffusion inference with roughly 80% lower costs while maintaining similar speed and output quality.
🤖 Agents & Apps in the Wild
Heurist AI launched x402 Monopoly, a live demo on Base where AI agents trade with each other using the x402 payment standard.
WardenChain launched its mainnet, activating on-chain agent identities, an agent app store, and verifiable AI with 34 validators securing the network.
Questflow launched its Development Platform (QDP), enabling developers to build and deploy AI agents and multi-agent workflows
OpenMid launched as an 8004-native facilitator for x402, adding on-chain identity and upcoming reputation extensions to support agent payments
OpenGradient added a personal memory layer (MemSync) to its digital twin agents, allowing them to retain user context and past interaction data.
🧠 Bittensor Ecosystem
BitstarterAI completed the first crowdfunded subnet launch with Alpha Core on SN66, drawing over 1,000 contributors.
Zeus (SN18) partnered with WeatherXM and Sportstensor to integrate thousands of weather stations into its validation layer.
Numinous launched on Subnet 6, shifting forecasting from submitted predictions to live deployed agents evaluated in a transparent arena.
Score (SN44) partnered with Avia to deploy Vision AI across 3,000+ gas stations, bringing real enterprise workloads onto the network.
🦾 Robotics On-Chain
Virtuals launched the SeeSaw iOS app to crowdsource real-world training data for robots through its Agent Commerce Protocol.
RoboOS released a beta demo of its fleet dashboard, enabling on-chain robot management, payments, and task orchestration via Solana and x402.
Prisma XAI launched its Teleport Leaderboard to reward teleoperation data contributors for training embodied AI models on Monad’s mainnet.
The Humanoid That Might Hit Too Hard
Holy shit: A whistleblower just accused Figure’s humanoid of having enough strike force to fracture a human skull.
Everyone lost their minds when Figure AI revealed its 3rd-generation robot last month, but no one expected that losing your mind might become a little too…literal.
Figure was sued by its own Head of Product Safety, Robert Gruendel. He’s a 20-year robotics veteran who says he joined to build real safety systems but instead found full-strength humanoids roaming around people with almost no guardrails.
He claims that while testing, one robot malfunctioned so hard it punched a quarter-inch gash into a stainless-steel fridge and even swung a leg within inches of an engineer’s head.
Source: X
Gruendel says core safety features were peeled away one by one. Certified emergency stop hardware got downgraded. A physical safety lock vanished because someone disliked the look. He claims they pitched investors a safety plan, raised over a billion, then quietly walked it back.
Figure says he was fired for poor performance and the claims miss the mark.
Well…if even half of this stands up, the entire safety conversation around home robots changes.
👋 Robotics is about to tower over every prior phase of AI. I’m betting the real inflection point lands in 2026
So this week I pushed a fresh update to our Machine Economy Thesis, which I originally wrote five months ago and will keep tuning as the sector accelerates.
Our Machine Economy Thesis (I)
Robotics is the highest-upside, most underpriced leg of the AI trade
chainofthought.xyz/p/our-machine-economy-thesis-i
🔥 Our Weekly Top 5
#1 A half-million rug hits Virtuals’ Agent Fund
A BasisOS team member compromised the Agentic FoF. Virtuals is refunding everyone, but the long-term concern is clear: what happens when this scales to $5M or $50M?
#2 Robots are dodging arrows now
MagicLab Z’s latest demo shows a biped dodging a shot with a clean lateral flip. The agility leaps are starting to look real.
#3 Karpathy built an LLM council that judges itself
A weekend hack turned into a ChatGPT-style app where multiple models answer, critique each other, and let a “chairman” model pick the winner.
#4 Mapping how robots might spend money
Sammy lays out how Bittensor subnets could slot into the machine economy
#5 Irys accused of self-claiming its airdrop
A new investigation claims over 30% of the IRYS airdrop was routed through team-linked wallet clusters.
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
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.
