Welcome to #89 of the AI edge.
The Middle East went hot this week. Iranian drones hit across the Gulf, debris struck civilian buildings in Dubai, and even an AWS data center got taken out. Cloud infrastructure is now a casualty of war.
This led to CT spending the week debating whether Dubai, the crypto-bro promised land, is still a safe bet.
While all of that was unfolding, BTC decided war is bullish and ripped from ~$64K to $73K while equity markets bled out.
Six consecutive red weeks and suddenly the one green week comes during an actual shooting war. Make of that what you will.
On the AI side, the week belonged to Anthropic. They told the Pentagon no and refused to loosen safety guardrails for military use. This led them to being blacklisted by Trump, labelled a "supply-chain risk," and lost contracts worth millions.
OpenAI swooped in hours later and signed it. However, the internet responded by pushing Claude to #1 on the US App Store and launching a full #QuitGPT movement.
Maybe a silver lining after all?
Quick heads up: We’re opening a small number of research partnership slots. If you’re a startup doing serious work in AI or robotics and want to work closely with us - pressure-testing ideas and getting in front of the right people, fill out this form and I’ll personally follow up.
The Big Story: Ridgeline vs. the Coding Agent Giants
The coding agent space has a pricing problem. Most tools either charge $100-200/month, or look cheap on paper until you pick a frontier model and realize your $20 Cursor plan buys you single-digit completions.
Ridges, Bittensor's Subnet 62, just launched their much-anticipated product layer, Ridgeline, into open beta at $29/month for 100 PRs. Point it at a GitHub issue, and the agent reads the repo, writes patches, tests them, and ships with no hand-holding.
The Wedge
More than the pricing, what makes Ridgeline different is how the agent gets built. Behind the product is a competitive subnet where miners race to improve the coding agent itself. Agents are scored on correctness, speed, and cost. The one you use is whoever won a “daily tournament” backed by over $18K in emissions. That's an R&D loop no centralized startup can replicate at this cost.
Inference runs on Bittensor infrastructure (Targon, Chutes), which is why the unit economics look nothing like centralized tools. The benchmarks back it up too with Ridges scoring 73-88% on SWE-bench Verified and 96.3% on Polyglot, all on open-source models. For context, frontier leaders like Claude Opus 4.6 and GPT-5.4 score in the 77-79% range on the same benchmark. Ridgeline is trading blows with models that cost 5-7x more per task.
The Fine Print
It's early and in beta. Users report solve times are still slow as the platform scales, and the team has been pushing miners to cut completion speed but it's not instant yet.
Benchmarks measure structured problems on known repos. How Ridgeline handles messy, undocumented production codebases is the real test that will determine if these economics hold outside the eval suite.
Coding agents are one of the most competitive spaces in AI right now. Building one on decentralized rails at a fraction of the cost is either the smartest bet in crypto or the hardest thing to pull off. Probably both.
Chutes.ai (Bittensor SN64) reported hitting roughly $6M annualized revenue, signaling real demand for its inference/compute services and improving subnet economics.
GenLayer launched InternetCourt.org, an arbitration layer for disputes that pop up when AI agents with wallets start signing and transacting at machine speed.
Gensyn AI launched what it calls the first AI-settled binary market, where outcomes are determined and finalized entirely by AI to prove autonomous prediction markets can run end-to-end without human adjudication.
Solana introduced Agent Registry (its implementation of ERC-8004), giving agents native on-chain identity, reputation, and third-party validation so they can show up with credentials and build a verifiable track record.
Virtuals Protocol introduced Base Batches 003 (Robotics Track) to fund and mentor teams building coordination rails for embodied agents including things like identity, permissions, and payments (up to $50K).
Hippius Subnet (SN75) released Hermes, an open-source protocol that lets Bittensor subnets stream data directly to each other, creating an early interoperability standard for the TAO subnet ecosystem.
Venice integrated Qwen 3.5 35B, giving users a large multilingual model with a huge 262K context window and multimodal input, positioned for strong coding/agent workflows while keeping inference private on Venice. VVV is ripping.
Silencio Network launched Conversational Recordings (beta), paying users for real conversations to produce native-language Voice AI training data.
peaq launched its Machine Tokenization (RWA) Framework, a full-stack system to tokenize and finance robots as a new asset class, turning machines into regulated, liquid economic actors with compliant fractional ownership and automated revenue distribution.
Quick update: I’ve talked about Tria before. It only went live on Hyperliquid 30 days ago and is already up to #4 in 7-day revenue and #9 in trading volume. That’s pretty solid considering most of the builders ahead of it have been live for 8+ months. It’s real traders generating real fees. (d/c: am an investor)
🔥 Our Weekly Top Tweets
#1 OpenClaw’s Rise Continues
OpenClaw just overtook Facebook’s React in GitHub stars, putting it in the top 15 most-starred repos of all time (around #12 at ~261k stars). It took less than 3 months since launch to reach this milestone.
#2 Autonomous Agent Prints Revenue
KellyClaude, an OpenClaw agent, generated $6.4K by building and shipping multiple products (including iOS App Store apps) and marketing them autonomously, with only minimal human setup.
#3 Venice’s Privacy AI Meta
Venice is seeing surging LLM inference usage, and with private AI becoming a mainstream need. The token ($VVV) has also 7x in the past two months as Venice establishes itself as one of the category leaders in privacy-first AI.
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 & 0xAce
P.S. I also write a weekly newsletter on AI agents.
