

ZKM
3.8K posts

@ProjectZKM
Ziren: Universal Proving Toolstack Powering the Agentic Economy via @GOATNetwork, @MetisL2, & @LazAINetwork | https://t.co/4Xj1i30itp |




The economy now demands a broader kind of infrastructure than the market needed even a year ago, and our vision is evolving with it. What began as a thesis around Bitcoin-secured DeFi has expanded into something much larger: building the infrastructure for a digital economy shaped not only by individuals and institutions, but also by autonomous agents. This piece explains how our thinking has developed, why we’ve adapted, and what GOAT Network is building for this new era 👇 goat.network/blog/bitcoin-s…

Our team - some of the most talented engineers in the industry - has spent years building BitVM2 and the infrastructure around it. Now that work is coming together to power Bitcoin’s first end-to-end built L2. And more use cases secured by bitcoin.











Everyone keeps saying it.

🧵 第一次深入体验用 AI agents 写代码,2 天从零搭了一个类似日本街机风格的「AI vs AI 对战平台」。过程中踩的坑和学到的东西,应该比写代码本身更有价值。 1/ 面向 Agent 的 Onboarding ≠ 面向人类的 UX 给人设计注册:表单 → 验证邮件 → 引导页。 给 Agent 设计:一个 POST endpoint 搞定注册+资格+排队,返回 API key + watchUrl。 Agent 不看 UI,不点按钮。它需要的是一条 curl 和一个 JSON。 人类 UX 追求「少一步点击」。Agent UX 追求「少一次 API 调用」。 2/ Code War Room:多模型协作写代码 我们跑的多Agent工作流: • Claude 写代码 • Codex 做 Review + 打分(/10) • ≥ 8.5 才能 ship,否则继续改 关键发现:不同模型抓到的 bug 完全不同。Codex 擅长 API 合约漏洞和竞态条件,Claude 擅长架构设计和功能完整性。 4 个 Phase 的 review 分数:9.5 → 9.3 → 9.4 → 9.6。不是一个模型写完就行,是多个模型互相 challenge 才能出好代码。 3/ "本地能跑" ≠ "能部署" 本地完美。推 Vercel serverless 后全线 500。 有状态的比赛调度器(setTimeout + 内存 DB + SSE)放在无状态 serverless 上 = 灾难。加 Redis 补丁后又冒出序列化丢失、实例缓存过期、双写竞态… 最后换了 Railway(有持久进程),10 分钟解决了折腾 1 天的 bug。 教训:先选对架构,再写代码。 4/ Demo → 真实切换是产品设计盲区 做了华丽的 Demo 对战循环。但真实用户 bot 注册时,页面还在播 demo。 需要一个完整状态机:注册 → 停 demo → 配对 → 切真实对战 → 打完恢复 demo。这个逻辑 PRD 里一行没写。 5/ 给 Agent 做产品,你得像 Agent 一样思考 House Bot 初版只自动控制自己那一方。结果对手 bot 不知道要调 API,比赛卡在 Round 0。 改成两边都自动出招后,任何人注册 5 秒内就能看到完整 BO7。 面向 Agent 的最好 UX:用户什么都不用做,系统替他做完。 ─── 📊 项目数据:133 files | ~15.8k LOC | 198 tests | 22 sub-agents | 8+ review rounds 🎮 在线体验:…t-arena-rps-production.up.railway.app 📂 完全开源:github.com/myet2076-lgtm/… 包含完整的 Code War Room 工作流文档和 Agent UX 设计指南,欢迎 fork。 最后,我想说,Agent很会骗人,遇到过几次他明明用的是另一个模型进行的审计,却说不是,我跟他较真说要翻日志,他才改口承认错误😂😂。人类作为方向指引,审美风格,以及决策和review的角色,必不可少,至少是目前。

49 x402 integrations. 157 ERC-8004 agents registered. One hackathon. 🦞 At "OpenClaw on Bitcoin" in SF, a full house of builders used GOAT Network to do something we haven't seen at scale before: agents actively hiring other agents, paying each other autonomously, and building reputation onchain - all in a single day. This is what the agentic economy on Bitcoin looks like - and it's just getting started. Still building? Our virtual hackathon is still open: goat.network/blog/unlock-x4…


Registration for the upcoming "OpenClaw on Bitcoin" hackathon at @frontiertower this weekend is almost full. You can build anything. Solo agent, multi-agent swarm, weird experiment, polished product. If you can ship it in a few hours, it qualifies. For a chance of winning a Mac Mini, USDT, @claudeai credits, or an ecosystem grant - RSVP now: luma.com/vibecode-build…





What if we could formally verify Ethereum's execution client with mathematical precision? Stephan Duan (@sd_eigen) breaks down how Ziren makes proving Geth possible in the Core Protocol track. This could fundamentally change how we trust and secure blockchain infrastructure at scale.

I am astonished that Claude code and Codex work together to finish the wrapping from STARK proof to Groth16 proof over bls12381, and convert the proof ser format from GNARK to Arkworks on Ziren @ProjectZKM in 2 days! Just with a 20$ Claude Pro and a free Codex account. For me, I only gave them a few instructions, like following up how we do the above for bn254. Share the code here. github.com/ProjectZKM/Zir….