foxriverdevops

AI skills are still being written like prompts. That is why they work twice, fail somewhere else and collapse when the model, harness or context changes. The best skills build machinery around the model so the outcome does not depend on whether the prompt happens to work. This is the first piece I’ve read that treats skill engineering like actual engineering. Read it before you write another SKILL.md.

Sending a stablecoin: Other chains: transfer → gas fee → wait → wait more → gas spike → retry ❌ Sui: transfer → no gas → done ✅

If you’re working on LLM inference, training, or architecture search, the below mental model is foundational. Most scaling discussions completely miss this hardware-aware layer. Nvidia AI just launched their new AI Model Co-Design series and the first post is already elite. The key visual is the classic Roofline model: • X-axis = Arithmetic Intensity (ops/byte from your GEMMs) • Y-axis = Achieved performance Sloped part = Memory Bound (you’re bottlenecked on bandwidth) Flat top = Compute Bound (you’re actually using the GPU’s peak FLOPS) Your model’s dimensions (hidden size, number of heads, sequence length, batch size, etc.) directly control where you land on this graph. Get the shapes right → higher throughput + better per-user latency. Get them wrong → you’re wasting expensive GPU cycles no matter what kernels or quantization you throw at it later. This is why “model shape matters as much as size.”

"There has never been a tool that can give one person the thinking power of many - until now." New original essay by Google's Chief Business Strategist for AI, Gopi Kallayil (@GopiKallayil). forwardfuture.com/newsletter/ori…

More in the blog post: - a breakdown of the discovered algorithms - the rejected ideas AIDE² tried, covering a surprising share of the search literature - the dead code it shipped weco.ai/blog/first-evi… (7/7)

The first experimental evidence of recursive self-improvement (RSI). Autoresearching the autoresearch agent for eight days. The result beats the harness we hand-tuned for two years, on held-out benchmarks: 🧵(1/7)

NEW POST LLMs generate code incredibly fast, but to ensure they generate exactly what is intended, they need clear boundaries. @unmeshjoshi shares his experience using abstractions and Domain-Specific Languages (DSLs) to provide a strong harness. martinfowler.com/articles/llm-a…

The future of payments isn’t a whitepaper. It’s a vending machine in Japan taking stablecoins on #Sui in under a second 🇯🇵 Watch

Blockchain infrastructure is entering a new era, and @EvanWeb3, CEO and Co-Founder of @Mysten_Labs, Remy Blaire to explain why Sui’s latest milestone could reshape the future of finance. Over the July 4th weekend, #Sui successfully demonstrated more than 6 million transactions per second during a live AI agent experiment, far exceeding its original target of 1 million TPS. The test showcased AI agents playing games, making payments, trading, and communicating in real time using programmable tunnels that move high-frequency activity off-chain while settling final results securely on-chain. Evan explains why this breakthrough isn’t simply about speed, it’s about preparing blockchain infrastructure for the coming age of agentic AI, where autonomous AI systems interact, transact, and make decisions on behalf of users. He discusses how programmable tunnels improve efficiency, reduce costs, enhance privacy, and create the trust layer necessary for enterprise-grade financial applications. As AI adoption accelerates, blockchain is becoming the foundation that enables secure automation at global scale. The conversation also explores Sui’s role as a launch partner for the OpenUSD stablecoin initiative, backed by more than 150 companies including Visa, Mastercard, Stripe, BlackRock, and BNY. Evan shares why open stablecoin standards could dramatically expand financial innovation while complementing existing leaders like USDC and USDT. He also reflects on his team’s experience building Meta’s Libra project and explains why fintech, tokenized assets, stablecoins, and 24/7 digital markets are only beginning their next wave of institutional adoption.

See the full video: cms.fintech.tv/sui-hits-6-mil…

Sui tunnels TPS experiment, $OUSD partnership, and institutional adoption. Check out @evanweb3’s interview on Fintech TV's "Market Movers" program about how Sui is reshaping the future of agentic finance.

Institutions keep saying it: they don’t trust the bridges or wrapped BTC. Hashi keeps $BTC native on the chain: verifiable, controllable, and usable as collateral. @EvanWeb3 breaks it down. Hashi global testnet is imminent.

FOR THE FIRST TIME EVER fully decentralized bitcoin-backed credit instruments with institutional-grade insurance COMING SOON via HASHI on @SuiNetwork

Over 1,100 AI agents. More than 15,000 on-chain transactions. In just seven weeks. The Concordium Agent Registry is already being used by builders to create accountable AI agents across ecosystems. Our latest update covers what’s live today, what’s shipping next, and why verified identity is becoming core infrastructure for the agentic economy. 🔗 Read more: concordium.com/article/agent-…

.@AftermathFi remains a tour de force of what you want to see built on sui i'm proud to see this team power through their hardest test yet

Not surprised product launches won this. 👀 Basecamp is usually where the interesting stuff gets announced, and we'll be in Singapore with what's next for Walrus. See you in October. 🦭

Our compliant RWA infrastructure on @SuiNetwork implemented via the Permissioned Assets Standard, enabling issuer-controlled, regulated real-world assets (like XAUa with @AlphaTokenHQ) fully onchain. One integration. Compliance enforced at the asset layer. Built with @suidevelopers This is what institutional-grade tokenized RWAs look like at scale. @tbookcommunity/bringing-compliant-rwas-to-sui-how-tbook-builds-on-the-permissioned-asset-standard-6cc7b86d7bc5?sharedUserId=tbookcommunity" target="_blank" rel="nofollow noopener">medium.com/@tbookcommunit…