GM {{ First_Name | }} !

I’ve been thinking a lot about personal compute lately. Not GPUs.

Our brains.

Right now, the best advantage we can have is maxing out our own mental capacity and ā€œprocessing powerā€. That means eating well, getting enough sleep, lifting, moving, staying mentally sharp. Nothing fancy. Just the basics done well.

That’s become even clearer with AI in the picture. Used well, AI is a massive force multiplier. It will 100x your mental capacity if you use it right.

Our minds are the bottleneck. Scale that and you're unstoppable.

Now… on to one of the most jaw-dropping AI stories of the year.

Windsurf AI, a breakout coding tool pulling in $100 million annually, was about to get acquired by OpenAI for $3 billion. Early employees thought they were set, surfing for new homes to buy on Zillow. Life-changing moment.

Then it blew up.

Microsoft, a major OpenAI investor, reportedly wanted Windsurf’s IP folded into the deal. Sam Altman pushed back. The talks collapsed. The offer expired. Just like that, the deal was off.

The next day, Google scooped up Windsurf’s CEO, execs, and top engineers in a $2.4 billion hiring raid. Everyone else was left behind. No acquisition, no payout. Just a broken deal and an empty office.

But then came a twist.

Cognition Labs, the team behind the AI software engineer Devin, stepped in. They’re buying the rest of Windsurf and keeping the staff on payroll, no matter how new they are.

A $3B dream collapsed, then got a second chance. Silicon Valley’s craziest plot twist yet.

Alright, let’s get into the rest.

BTW: We’re looking for a sharp, curious researcher to join our team. Think you’ve got what it takes? Apply here.

Chainbase launched $C, its DataFi token. Airdrop claims are open through July 28. The token covers data access, network security, governance, and incentive mechanics across the platform.

Warden added WachAI to its Agent Hub, a due diligence agent that gives token risk assessments and cross-chain data with a single prompt.

Aspecta, which focuses on bringing liquidity and verification to illiquid assets, raised funding from YZi Labs to help scale its network and adoption.

šŸž Fresh alpha from our end:

  • Watch Frodobots closely, there’s interesting stuff brewing.

  • CredMont, a data DAO from Vana, has been flying under the radar. But it won’t stay quiet for long.

  • We’re also seeing more chatter around Codec, which is building vision-language models aimed at robotics.

And in Web2 land…

The foundational AI model companies are printing. Anthropic jumped from $1B to $4B in annualized revenue in just six months. OpenAI is pacing at a $10B run rate, nearly twice its 2024 close.

Source: x.com/deedydas

And it’s not just the model labs. AI apps are scaling at speeds we haven’t seen before. Cursor, a dev tool built around AI, hit $100 million ARR in less than a year.

Call it what it is: AI growth is in overdrive.

⛺ Camp’s Testnet Crosses 1M Daily Txs

Camp Network is cooking, now processing over 1 million transactions a day on its testnet.

It’s a new blockchain built to give creators more control over how their work is used in AI systems.

With Camp, you can register things like music, code, images, and even personal data on-chain. You decide how it can be used, who can modify it, and what someone owes if they use it.

The core system is called Proof of Provenance (PoP). It records where content came from, how it’s used, and who should be credited or paid. It’s designed to prevent unauthorized scraping and help enforce usage terms automatically.

The basic flow:

  1. You register your work through Camp

  2. You set the rules, who can use it, remix it, and how much they pay

  3. AI agents or apps can use your content only if they follow your terms

  4. Payments happen automatically

  5. Everything is recorded on-chain

Source: docs.campnetwork.xyz

Camp’s testnet has already handled 55 million transactions from 4.3 million wallets, which is…a lot of activity for a protocol that’s still finding its footing.

Developers can spin up their own isolated chains on top of Camp. That’s helpful if your app needs to move fast without waiting in line, especially useful for AI tools that need constant, high-volume access to content.

There’s also mAItrix, Camp’s built-in agent framework. It helps developers build AI systems that can learn from registered work while respecting who made it and how it’s allowed to be used.

At its core, Camp is trying to make digital ownership real again. So AI doesn’t just use content, but it pays for it.

🌱 GaiaNet Hits 700,000+ Active Nodes

That’s a massive number, putting it among the biggest decentralized AI networks in existence.

The concept is simple: anyone can create their own personal AI agent, trained on their skills, preferences, and workflows. GaiaNet describes it as ā€œYouTube for knowledge,ā€ but it’s really more like a digital assistant that works the way you do.

Source: docs.gaianet.ai

Each node runs a full AI stack. That includes a fine-tuned language model, private vector search, long-term memory, access to APIs, plugins, and function-calling tools. These agents can search for information, plan tasks, call real-world APIs, and act on the results. All of it runs locally, on your device.

Setup takes a single click. If your node stays online, you earn rewards. Payments happen through Purpose Bound Money smart contracts. Users prepay for tasks, and nodes get paid automatically once the work is done. The GaiaNet token handles all payments, governance, staking, and rewards for reliable operators.

Everything is open-source, privacy-focused, and designed so anyone can run it. No cloud hosting required.

Protocol Updates

  • Vana’s rolled out its Capella upgrade. The update improves throughput, tightens security, and adds better tooling for building Data Apps and Liquidity Pools. It’s meant to make user-owned data easier to work with and more secure.

  • Nous Research open-sourced Hermes 3, a dataset with over 1 million high-quality examples built for training uncensored, roleplay-capable AI agents. It helped push Llama-3 models to new performance highs.

  • Phala launched a cloud GPU platform with trusted execution (TEE) support. You can now deploy models privately, rent GPU time, and use OpenAI-style APIs, all with cryptographic verification.

  • Mira released Verify, a fact-checking tool that uses multiple AI models to flag misinformation online. It tags claims as true, false, or uncertain based on model consensus. Basically, it’s multisig for truth.

  • Cysic, known for zero-knowledge hardware, joined the Succinct Prover Network to help speed up proof generation. They’re contributing both custom hardware and protocol-level experience to boost performance.

  • Nexus kicked off Community Point Incentive 1.0. Testnet III contributors can now earn points for building, creating, or participating. It’s a lead-up to mainnet, with early community rewards in play.

  • Pond is running a 14-day virtual hackathon with support from Google Cloud, Eleven Labs, and the Ethereum Foundation. The focus is on helping early AI projects ship MVPs, get funding, and find users.

Bittensor

  • Yuma now runs a validator that gives users direct access to Bittensor ($TAO) staking through Crypto.com. It lowers the barrier for anyone who wants to participate without setting up their own infrastructure..

  • Ridges AI released a browser-based IDE for Bittensor miners. It includes code analysis, performance metrics, and logic flow diagrams to help developers debug and optimize their miners more easily.

  • Manifold Labs launched the Targon Order Book, a system where miners can list ask prices for their H200s and earn based on market demand. It’s a first step toward open, auction-based pricing for compute resources.

AI Agents

  • Teneo launched its AI Agent Testnet. Browser nodes now act as on-chain AI agents that process live public data. Access is limited for now, but Golden Ticket holders are already in.

  • Crestal Nation released an API for adding AI agents to real applications using standard HTTP calls. It includes built-in support for messaging, memory, and OpenAI models. Think of it as a DIY toolkit for running autonomous AI inside your app.

  • Giza launched Arma on the Base app, a personal AI agent that lives in your wallet and handles asset allocation 24/7. Just chat to earn stablecoin-native, risk-adjusted yield on Base.

  • ACT LabsĀ introduced ACT Flow, a shared workspace where people and AI agents complete tasks together, verify outputs, and earn in a transparent, on-chain system.

Web2 AI

  • The big one this week: Moonshot AI dropped Kimi K2, a large open-source model built for code and agent tasks. It leads on coding benchmarks like SWE Bench and Tau2.

  • LangChain open-sourced Open Deep Research, a LangGraph-based framework where sub-agents collaborate to build structured, in-depth reports. It supports custom models, tools, and server setups.

  • AWS dropped Kiro, an AI IDE that helps devs go from prototype to prod using specs and smart hooks. It blends agent-speed coding with actual structure so your app doesn’t melt in prod.

  • Mistral launched Voxtral, a new set of open-source speech models that outperform Whisper, GPT-4o mini, and ElevenLabs. Features include multilingual transcription, voice-based Q&A, long audio support, and function-calling.

  • Liquid AI introduced LEAP, a toolkit for running compact LLMs directly on mobile devices. They also released Apollo, an iOS app for testing models locally.

You can’t trust AI.

But what if it had to back up its answers?

In our latest research deep dive, we break down how verifiable AI might actually work, and take a closer look at how big this market could get.

Worth reading before the next model lies to you with a straight face. šŸ‘‡

šŸ”„ Our Weekly Top 5

OpenAI just killed another 100 AI agent startups with its new feature release

Spheron Network hits $10M ARR ahead of their TGE

Clearest path to a $10M ARR B2B AI Agent startup

xAI launches Companions on the Grok app. Don’t ask us what NSFW things you can do on it.

Bossman Yann LeCun says: If you’re interested in human-level AI, don’t work on LLMs

That’s a wrap for this week! Got thoughts, feedback, or something cool to share? Just hit reply. We read it all.

🚨 Quick heads up:

  1. Are you the telegram yet? We drop daily AI alpha that you don’t want to miss.Ā Ā 

  2. Catch our podcast summaries, complete with detailed show notes.

  3. And if you want to jam with us directly, hop into our Discord

Cheers,

Teng Yan & Issy

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.

Reply

or to participate

Keep Reading

No posts found