News
Curated articles, my blog posts, and site announcements.
Curated articles, my blog posts, and site announcements.
Curated articles, my blog posts, and site announcements.
Legato is a private payroll system on Tempo where employers deposit salaries into a privacy asset pool and employees withdraw privately at any time. Built with Noir, Solidity, and the Tempo testnet, with the source fully open-sourced on GitHub.
Aztec co-founder Zac Williamson argues that privacy on top of centralized rails turns the settlement layer into a permissioned service with platform risk, viewing keys, and approval bottlenecks. Real programmable private state must be decentralized from the ground up: users execute locally, contracts compose privately, and disclosure stays at the application layer through selective disclosure rather than network-level backdoors.
Merces is TACEO's core protocol for financial privacy on shared state, supporting payments (including x402), yield, and tokenization with built-in compliance enforcement and configurable audit mechanisms. Verifiability comes from ZK, while composability and privacy are built on MPC and ORAM.
Aztec Network has gone fully permissionless after onchain governance revoked ownership of the core rollup smart contract, making its code and config immutable. Governance delay drops from 30 days to 2 days; users keep an escape-hatch exit. Aztec Labs has flagged proof-system vulnerabilities that can only be fixed in a future rollup version.
Nous Research's Hermes Agent now ships a /learn command that ingests directories of source material, code, API docs, manuals, PDFs, configs, and distills them into verifiable, reusable skills that the agent can call back into later.
Jason Yuan's HIVEMIND has launched, opening a closed beta program for a new social intelligence experience. The reveal film is built around true stories from early users including Biz Stone, and the team has published a constitution outlining the values governing the network.
Aside, a YC F25 AI browser, is now live with vertical tabs and Liquid Glass styling, SOTA results on agentic browsing benchmarks (Online-Mind2Web 297/300), fully local + encrypted execution, and support for using your own Claude or ChatGPT subscription.
Modular has entered an agreement to be acquired by Qualcomm, accelerating its push to unify accelerated compute behind an open software platform. The deal also marks Qualcomm's evolution into a developer-first AI solutions company spanning generative and agentic AI from edge to cloud.
OpenAI has designed and built its first AI chip, Jalapeño, brought to production with Broadcom and purpose-built for the LLM workloads powering ChatGPT, Codex, the API, and future agentic products.

Zenith has been invited to join Progmat's Tokenized JGB and On-chain Repo Working Group alongside MUFG, Mizuho, Sumitomo Mitsui, BlackRock Japan, State Street, SBI Securities, and Japan Exchange Group, conducting a joint study on bringing Japan's $1.6 trillion JGB repo market on-chain. A comprehensive report is expected in October 2026 with issuance pilots targeted for later this year.
Allium has raised a $40M Series B led by Amplify Partners, with continued backing from Kleiner Perkins and Theory Ventures. Since their Series A, the onchain data platform has grown revenue 10x, scaled to 150+ enterprise customers, and now indexes 30+ petabytes across 150+ chains.
UBS and Nethermind have completed two joint proofs of concept showing that a public, permissionless network can support the compliance and operational needs of regulated financial institutions. They built a node that applies customizable compliance rules to outgoing transactions and a routing component that sends approved bundles directly to selected block builders, both validated end-to-end on Sepolia.
MoonMath AI has soft-launched a private AI coding serving endpoint, powered by MiniMax M3 and running on its own EU-based GPUs. The pilot controls the full software and hardware stack with no personal data collection and no prompt retention, priced significantly below closed-source model APIs.
The Ethereum Foundation is cutting its budget by roughly 40% as it transitions to a long-term endowment-based organization, shifting from a pre-2026 average of ~15% annual spend to a post-2030 target of ~5%. Sacrifices include moving the multi-client model from redundancy toward specialization with AI-assisted formal verification, winding down PSE as a unit, smaller-scale Devcon, and fewer beyond-Ethereum megaprojects.
Cursor unveiled three announcements at its inaugural Compile conference keynote, including a new model being trained in partnership with SpaceX.

The Ritual Chain public testnet is live. Built specifically for AI agents, it offers on-chain inference, composable precompiles (LLM inference, ONNX models, HTTP calls, DKMS), and native scheduling, letting smart contracts call an LLM, verify the result, and act on it entirely on-chain.
OpenAI is expanding Daybreak to democratize patching vulnerable software at machine speed: a Codex Security plugin for finding, validating, and fixing vulnerabilities inside Codex; the full GPT-5.5-Cyber model for trusted defenders; a Cyber Partner Program for security companies; and Patch the Planet, an effort with Trail of Bits, HackerOne, and maintainers to secure critical open source projects.
Ethlabs launches as a non-profit R&D lab with a mission to make Ethereum the settlement layer of the global economy, working between frontier builders and protocol development. Backed by a broad coalition of DeFi builders, core devs, L2 founders, cypherpunks, investors, institutions, and researchers.
The Ethereum Foundation lays out its execution mandate: eliminate toxic MEV, make unconditional privacy the default, treat staking as protocol infrastructure risk, harden every surface against capture, and refuse to compromise Ethereum into a compliant fintech rail with web3 branding.
Valar Atomics took its Ward 250 test reactor critical for the first time, fulfilling EO 14301 which called for 3 advanced reactors to go critical by July 4th. Ward 250 is a TRISO-fueled HTGR and the first nuclear reactor ever air-lifted by a C-17. Power operations targeted for July 4.
Enshrining a native zkEVM is a core pillar of Ethereum scaling. Combined with Blocks-in-Blobs (EIP-8142) and Data Availability Sampling, zkEVM lets gas-limit increases have sub-linear impact on block download latency, scaling both execution and bandwidth.

EthLabs is a new non-profit R&D lab focused on making Ethereum, ETH the asset, and the ecosystem win. With EF on CROPS, EthLabs on R&D, and Etherealize on institutional adoption, the ecosystem now has clear stewards across the stack.
World App now lets users receive paychecks directly in USDC, powered by Bridge (a Stripe company), with self-custodial deposits and no requirement for employers to understand blockchains or manage network fees.
Changing arxiv to autoarxiv in any CS paper URL deploys an agent that resolves codebase setup issues, runs a minimal reproduction, and estimates full replication cost.
a16z New Media turns one. Erik Torenberg breaks down the four motions of the team: in-house creative (video, design, editorial), owned channels (X, Substack, podcasts), dedicated full-service team from pitch to launch, and a network leveraged with software and AI. Framed as go-direct as a service for portfolio companies.
A new draft EIP (PR #11776) proposes a Warm-Access Multiset (WAM): a sliding 256-block refcount of accounts and storage slots seen in Block Access Lists, so recently-touched items start warm in subsequent blocks. Empirical analysis cites ~14.1% median per-block gas savings, with a SHA-256 Sparse Merkle Tree commitment added to the block header for zkEVM-friendly inclusion proofs.

MegaETH introduces MOSS, a wallet permissioning layer for AI agents. The MOSS CLI grants scoped permissions (allowed contracts, functions, spend caps, time limits); the agent never touches keys and funds stay in user custody.
Midjourney is spinning up Midjourney Medical, a new division focused on applying its generative imaging stack to healthcare.
Midjourney walks through the technical architecture behind its new Midjourney Scanner.
GLM-5.2 ships with open weights, significant gains in coding and agentic tasks, a 1M-context window for long-horizon work, and two reasoning effort levels (max and high) at the same API price as GLM-5.1.
News tools
One compact deck for archive access, subscription controls, and a tighter recommendation list.