What Kaspa's Toccata fork actually unlocks The kaspa:native upgrade adds covenants, programmable rules attached directly to transactions, enabling vaults, escrow, and smart wallets without giving up proof-of-work or BlockDAG speed. KRC-20 tokens now live natively in consensus rules instead of inscription workarounds. New ZK opcodes let rollups and bridges anchor to Kaspa, with two EVM-compatible Layer 2s already building on top.

Shepherd：让 meta-agent 能像 Git 一样操作 agent 执行历史 当 agent 系统处理更复杂任务时，越来越依赖 meta-agent——创造、操作和管理其他 agent 的高阶 agent。Meta-agent 需要协调 agent、在危险动作执行前叫停、修复失败的运行——这些操作要求运行时能操控 agent 的执行过程。 但现有 agent 基础设施暴露的只是转录文本和环境快照。Meta-agent 被迫自己拼凑工具来重建和操作完整执行状态。 Shepherd 的解决方案： 把 agent 的每次执行本身做成一等对象。每次模型动作、工具调用、环境变化都变成可逆的、类 Git 执行 trace 中的一个结构化事件。任何过去状态可以回退——比 docker commit 和 fork 快 5 倍。 Python 实现，函数式编程基础。 开源。 三个用例的数字： 监督 meta-agent： 防止并行编码 agent 之间冲突，CooperBench 上的结对编码通过率从 28.8% 提升到 54.7%。 反事实优化 meta-agent： 修复 agent 工作流——提出修改并在改变行为的时间点开始重放运行。Terminal-Bench 2.0 上比 MetaHarness 高 12.8%，墙上时间少 58%。 训练 meta-agent： 在 rollout 期间选择分叉点来改善长时间 horizon agent RL 中的信用分配。在 Terminal-Bench 2.0 上让 GRPO 的提升翻倍。 "Git for agent execution"不是比喻——是真的。每一次 agent 运行都可以分支、回退、对比。这对生产中的 agent 调试和自我改进来说，可能是基础设施级别的基础件。 论文: "Shepherd: Enabling Programmable Meta-Agents via Reversible Agentic Execution Traces" arXiv: arxiv.org/abs/2605.10913 #MetaAgent #Agent基础设施 #RL

A programmable cache plus Workers Static Assets means that scaling up gets even cheaper.

Im a little surprised no ready options exist for simulation. What's your precise goal? Stable flight? Programmable motion? Self steering? If this was a project I was doing, knowing nothing about flight, here's the general shape of how I'd give it a shot with AI: Capture data set from real controlled flight, sensors, input, output, as much as you can get. Use this to train a model or heuristics so that you can predict eg sensor state from input. You have gyros or something I hope? Iterate here as needed, as you find weak spots. Keep collecting data. If you have specific goals (navigate some environment, etc) you'll probably need more pieces, of course, this is just basic control loop stuff. Once you can do this reasonably well, you now have a rough proxy test for actual controlled flight, and can iterate on controller via simulated flight. Give codex specific quantifiable hill to climb in sim (stable for X time, navigate these environs, etc.) If results work well in sim but not in real flight, work on your data pipeline some more, improve sim. But I'd spend time looking for existing sim resources first. This seems like a problem that's others will have solved at least some parts of already.

All of reality is programmable. You just have to figure out how. And the way to do that is to model it.

Bien dit ! Construis des revenus et des échanges en dehors des PSP, et développe une résilience réelle qui rendra ton adhésion au rail programmable optionnelle plutôt qu’inévitable..

Excited to share that Prova has been selected as a winner in the @FhenixIO Privacy-by-Design Buildathon on the @akindo_io platform. We're building the programmable infrastructure platform for modern trade credit, enabling insurers and trade finance providers to launch digital trade credit products without rebuilding the underlying infrastructure. Our focus over the next 4 weeks: • Customer discovery with insurers and trade finance providers • End-to-end platform validation • Pilot conversations • Shipping improvements based on customer feedback Thanks to the @FhenixIO and @akindo_io teams for the support. Back to building.

Agentic commerce benefits from a currency that is instantly transferable, programmable, and usable without traditional banking. #EGLD has the potential to become money for agentic commerce. It enables AI agents to transact, settle payments, and coordinate economic activity.

Digital Slaves: Is “Slavery” Becoming Ethical and Legal for the First Time? For thousands of years, slavery has always meant the same thing: One human exercising complete control over another. That is why slavery is not only illegal in most of the world. It is also one of the darkest chapters in human history. But what if we are entering a completely different era? For the first time, the “worker” is no longer human. AI models. AI agents. Soon, humanoid robots. They can work 24/7. They don’t ask for breaks. They don’t demand salaries. They don’t unionize. They don’t retire. They don’t get tired. What we’re building is a new kind of programmable digital labor. This raises an uncomfortable question. Could “slavery,” at least as a metaphor, become ethically and legally acceptable for the first time? Because no human is being enslaved. And based on everything we know today, these systems are not conscious beings that experience suffering or seek freedom. If they are ultimately tools, what is the ethical problem with running them continuously? The Industrial Revolution automated physical labor. The AI Revolution is automating cognitive labor. Tomorrow’s companies may not scale by hiring thousands of employees. They may scale by orchestrating millions of digital workers. The role of management itself could fundamentally change. Perhaps the real question is this: Was slavery always immoral because it involved absolute control? Or because the subject of that control was a human being? If AI ever becomes conscious, this entire discussion changes overnight. Until then, we may be witnessing the emergence of something humanity has never encountered before: A form of “digital slavery” that is legal, and perhaps, ethically defensible. What do you think?

Your point about programmable justice for AI agents is a crucial insight.

Human beings can tell the difference between slop and relevant data so the predictions markets are still sort of working. It’s the algo-saturated futures markets that are “programmable”. The future is not The AI Singularity - it is perma-GIGO. 🤖

プログラマブルロジックコントローラという長たらしい故障が正式名称なんだろうけど、俗には「シーケンサー」と呼ばれてるよね。昔のリレーシーケンスだとかを電子回路でやる機材だから。 CNCの制御装置の中にも入っていてPMC(Programmable Machine Controller)とか呼んでいたりもする。

TL;DR - Three wrong assumptions about blockchain privacy: 1️⃣ The primary buyer is the end user. In reality, most users don’t "care" about privacy, they seek better products. The strongest demand comes from institutions, where confidentiality is a prerequisite for bringing real capital onchain. 2️⃣ Privacy is about selective disclosure. Selective disclosure works well for traditional banking, where information could be shared with specific parties. But blockchain is programmable and composable. Onchain finance requires something different: the ability to compute on encrypted data, so assets can be traded, settled, or earn yield without ever revealing sensitive information. 3️⃣ Compliance should be built into the protocol. Compliance is context-dependent, while protocols should remain neutral infrastructure. Compliance belongs at the application and business layer—not hardcoded into the protocol itself. Think about it this way: you agree to the privacy policy of the application you use, not the underlying cloud provider. The GTM lead at Zama, Antoine (@Mr_Goodbuy ), expands these ideas in the article below ⬇️

Sixty-two holograms at once: a 6,144-element programmable metasurface projected independent images simultaneously, rather than one by one, hinting at far higher holographic capacity from a single input frequency. @NatureElectron techxplore.com/news/2026-07-…

🧵 SCB just became the first Thai bank to use Citi Token Services for 24/7 tokenized USD payments. 🇹🇭 But this raises a bigger question… Why is Thailand becoming one of the world’s most advanced blockchain banking markets? The answer isn’t just Citi. It’s an entire financial ecosystem. 👇 For years, the Bank of Thailand has taken a very different approach from many Western regulators. Instead of resisting blockchain… It created regulatory sandboxes. Banks were encouraged to test new financial infrastructure under regulatory supervision before launching it commercially. That changed everything. One of the biggest examples? @scb_thailand Thailand’s oldest commercial bank. SCB didn’t simply experiment with blockchain. It worked directly with the Bank of Thailand on: • Stablecoins • Retail CBDCs • Tokenized payments • Virtual banking Each project moved from controlled testing into real-world deployment. One milestone came in 2024. SCB and @lightnetgroup completed the Bank of Thailand’s regulatory sandbox for blockchain-based stablecoin remittances. The service was approved for commercial deployment using Fireblocks for digital asset custody. That wasn’t speculation. It was regulated financial infrastructure. Lightnet itself wasn’t built as a consumer crypto platform. It was designed to modernize cross-border settlement. Lightnet utilizes the Velo Protocol as its blockchain settlement layer for cross-border payments across Asia-Pacific. Now comes the latest announcement. @scb_thailand has become the first bank in Thailand to deploy Citi Token Services for 24/7 tokenized USD payments. Different technology. Different platform. Same direction. Notice the pattern? The Bank of Thailand. @scb_thailand @lightnetgroup @veloprotocol @FireblocksHQ @Citi They’re not building one single product. They’re modernizing different layers of the same financial system. Some focus on tokenized deposits. Others on stablecoins. Others on blockchain settlement. Others on regulated banking. Each solves a different piece of the puzzle. That’s why @veloprotocol remains relevant. Not because Citi suddenly uses Velo. But because the same market that embraced regulated stablecoin settlement through Lightnet is now embracing tokenized institutional payments through Citi. Both developments point toward programmable financial infrastructure. The question is no longer: “Will banks adopt blockchain?” Thailand already answered that. The real question is: How will tokenized deposits, stablecoins and blockchain settlement eventually work together? Thailand may be one of the first countries showing what programmable financial infrastructure actually looks like. Not replacing banks. Upgrading them.

Catherine Austin Fitts warns that programmable money, digital ID and AI-driven mass surveillance form three interlocking pillars of a digital control grid, enabling authorities to centrally enforce desired behaviour by "threatening to turn off your money unless you do exactly as you're told". "You're talking about moving into a slavery system where... your life can be controlled by people who make the rules very centrally, and enforce them centrally... stripping you of any constitutional rights or property rights."

SEDA's programmability will become the plumbing behind programmable finance. Whether deployers on @Perpsdotfun or the custom build outs for tokenized funds, CLOs , lending and RWA finance in general.

AI infrastructure isn't only about compute. The next bottleneck is moving massive amounts of data, fast, across clouds, systems, and machines. That's why we're building Beam on #Bittensor Bandwidth is becoming programmable. Thank you @YumaGroup ... See you July 7.