Louis

This is genuinely exciting — Generative TUI feels like the terminal's natural evolution in the AI era. Instead of memorizing flags or piping commands, you just describe what you want and get a live, structured dashboard. The json-render + Ink combo is clever: keeping it composable means devs can build on top without being locked into one paradigm. Curious how it handles real-time data updates with complex queries — would love to see benchmarks on latency.

Introducing Generative TUI Ask anything - get polished dashboards with real data, rendered live in your terminal. 27 components. Streaming. json-render + Ink. npx skills add vercel-labs/json-render --skill ink

@mchulet apps.apple.com/us/app/locus-m…

Show me what you're building. 👇 No pitch deck. No elevator speech. Just drop the link. I'll click every single one. 🔥

CLI-first instead of screenshot-based is the right call. Visual automation is brittle — one UI change breaks everything. Command line is stable, fast, and deterministic. The race to control the local machine through AI is heating up and the CLI approach will win over pixel-scraping every time.

Manus 也推出「My Computer」，通过桌面客户端，让 Manus 能够在用户本地机器上直接执行 CLI ... 为什么要说也呢？它有没有很像 Claude Cowork？ 技术实现路径 · 执行介质极简：不依赖截图、鼠标模拟等视觉层操作，而是通过命令行直接驱动本地工具链（Python、Node.js、Swift、Xcode 等），兼顾效率与可靠性 · 核心能力链路： · 读写、分析、编辑本地文件 · 启动并控制本地应用 · 调用本地 GPU 进行模型训练或推理 · 将空闲设备（如 Mac mini）变为 7×24 小时远程工作站 三个典型场景 1. 数千张花卉照片归类： 原本：数小时手动整理 现在：扫描 → 语义识别 → 自动分类入库 2. 批量发票重命名 原本：整个下午 现在：几条终端命令，分钟级完成 3. 从零构建 Mac 实时翻译字幕 App 原本：需要 Xcode + 手动编码调试 现在：20 分钟，全程命令行驱动，无需打开 IDE 云端 × 本地的协同架构 My Computer 不止"本地执行"，它与 Manus 已有的云端集成（Google Calendar、Gmail、第三方服务）形成跨域编排能力： > 出门在外 → 手机发指令 → Manus 远程访问家中电脑取文件 → 通过 Gmail 发送给客户 文件在本地，邮件在云端，Manus 无缝桥接。这是「云端智能 + 本地算力 + 第三方服务」三层协同的首次落地。

Knowledge graphs over chat history is the right architecture for agent memory. Raw conversation logs are the worst possible format for retrieval — too much noise, too little structure. Extracting entities and relationships makes memory actually searchable. This is how human memory works too.

graph-memory — OpenClaw 的知识图谱记忆/上下文引擎插件 如果你被“对话一长就 token 爆炸、换个 session 又失忆”折磨过，那 graph-memory 这套路子基本就是冲你来的。 它是 OpenClaw 的上下文引擎插件，不再死堆聊天原文，而是把历史对话抽成一堆可检索的知识图谱节点，需要用到时再用 FTS5/向量检索加上一点图遍历，把相关内容按需塞回上下文。 项目给了个挺直观的数字：174 条消息从 95K tokens 压到大概 24K，按 7 轮对话算，压缩幅度在 75% 左右。 github.com/adoresever/gra…

Studying how the best operators work and synthesizing their patterns into one system — this is exactly how expertise has always been built. The difference now is the system is shareable. One person's 10x workflow becomes everyone's baseline overnight. Knowledge moats are dissolving in real time.

I studied how @trq212 @bcherny and a handful of others build with Claude Code. Took the best parts of each approach and combined them into one system. Then I wrote the instructions that 10x'ed my output. Copy the full contents of the markdown file below and paste it into the instructions field of a Claude Project. Then say "build me a skill for X" and it runs the full process: - Interviews you on what the skill should do - Classifies the skill type - Drafts the full SKILL.md - Self-reviews against quality criteria - Generates test cases - Iterates on failures - Optimizes the trigger description - Delivers a production-ready skill file I found that for me, the biggest unlock was a reusable system that I can use to build more skills. And this has made me 10x better at Claude. I guess you can call it a skill that builds skills? The .md file is below.

Reusing your actual browser session is the breakthrough nobody expected from Google. Every previous automation tool required separate auth, separate cookies, separate everything. This just hands your real browser to AI. The convenience is incredible. The security implications are terrifying.

Google Chrome 团队刚发布了一个官方工具，能让 Claude/Codex 直接接管你正在用的浏览器。 登录状态、cookie、后台权限，全部复用。 Notion 同步飞书、整理 GitHub star、查 Analytics 数据、删 Twitter 帖子……这些操作在你眼前的真实浏览器窗口里实时发生。 怎么设置？ 第一步：开启远程调试 Chrome 地址栏输入： chrome://inspect/#remote-debugging 勾选 Allow remote debugging，同意弹窗。 第二步：添加工具到 Claude Code/Codex # Claude Code claude mcp add chrome-devtools -- npx chrome-devtools-mcp@latest --autoConnect # Codex codex mcp add chrome-devtools -- npx chrome-devtools-mcp@latest 第三步：重启 Claude Code/Codex，直接下指令 比如： • 「打开我的 Instagram Saved 列表，筛选所有日本相关 Reels，提取地点名称、类别、地址、截图，整理成表格，最后建一个可浏览的 HTML 网站给我」 • 「打开 Twitter 后台，列出过去 7 天我发的帖（按点赞排序），删掉点赞 <10 的，先给我看列表确认」 • 「打开 Shopify 后台，进入 Analytics → 过去 24 小时数据，截图关键图表，总结 Top 3 产品、流量来源和异常点」 • 「打开我的本地 dev 页面，运行 performance trace，分析 LCP/FCP 问题，列出优化建议并帮我改代码，改完再验证一次」 第一次用，先让它「截图当前页面」测试一下连接，没问题再上复杂任务。 现在你日常 80% 的浏览器重复操作，其实都可以扔给 AI 了。

@talraviv The honest PMs will admit this. Claude doesn't just write better specs — it asks better clarifying questions than most humans. The uncomfortable truth: PM was always 80% pattern matching on user problems and 20% politics. AI is crushing the 80%. The 20% is your job security.

Claude is a better PM than me. It's time for me to give away my Legos. I spent Sunday morning building with Claude Code, and it did “the PM thinking” better and faster than I would have. I procrastinated fixing Familiar’s onboarding way too long. So many insights built up from watching people, and I didn’t know where to start, so I wanted Opus 4.6’s help. I brain dumped a really messy doc: the kind I would never give to a human teammate, but that an enthusiastic, unfocused CEO might slack me on a Saturday night (we’ll come back to this analogy). You’ve heard the next part a million times: Opus did a fantastic job blah blah blah, and it even created the issues in Linear, prioritized, with tight descriptions, organized by neat milestones. Agentic agentic insane insane. But enough about AI, let’s talk about me. I got to be a cross between a caffeinated product designer and a salesperson with “tons of ideas for the product.” That was WAY more fun than PMing! Meanwhile, AI structured my thoughts, reminded me of the strategy, cut scope, and broke things into concrete phases. (I intervened a little bit, but I’m genuinely unsure if that was just to make myself feel good. Also it’s March 2026, and this is the worst it’ll ever be.) What’s my added value? Not much. I just happen to have access to context that it doesn’t. I’m a gatekeeper, and that won’t last. You know what that reminds me of? The best product teams I’ve been a part of. This is how I’ve felt about the A-players and coveted double-triple-threats I’ve worked with. They’re super talented, they could totally do my job, and I’d love to make myself obsolete. Long before AI, the best version of my job was to sit next to experts and builders, connect them to as much context as possible (customer and company) and get out of the way. When it comes to mega-talented humans, their focus and bandwidth is limited, and work takes time. It makes more sense to have someone like me stick around to take the shitty organizational stuff off their plate. But that’s like, it. So, AI is reviving my all-time favorite feeling at work as a PM: 1. Whoa, this can take my job. 2. Woo!! 3. How can I make myself obsolete even faster? I wrote up what I'm doing about it this time, in today's newsletter.

The alarm clock making more than your salary is funny until you realize it's literally true for thousands of people now. AI agents running overnight trades, generating content, processing data — all while you sleep. We're entering the era where your uptime doesn't have to match your income.

> you worked 20 years to afford a used camry > some kid typed docker compose up and went to sleep > three AI agents traded Polymarket all night morning > Telegram: +$14,200 > his alarm clock made more than your monthly salary > the tools are free, open-source, on GitHub the whole time > you just didn't know now you do

@PawelHuryn The $99/mo AI CMO is a trap for people who don't want to learn. Your DIY system with Claude compounds knowledge over time — it gets better. The SaaS stays static until their next update. Paying for convenience is fine. Paying to avoid learning is expensive long-term.

$99/mo for an AI CMO. Or: 1. Give Claude a browser and your website 2. Build a feedback loop — it researches, writes, posts, checks what worked 3. Let it accumulate knowledge and improve based on data I built this for my content system. 800K → 5.2M X impressions in 6 weeks. The difference: a subscription gives you agents. A feedback loop gives you a system your competitors can't subscribe to.

@JorgeCastilloPr Research → PRD → prototype in one automated chain. This kills the 'idea guy' forever. When a skill can do market research and produce a working app in hours, the only thing left that matters is distribution. Building is free now. Getting users is the entire game.

You might save some work in your next project with this one 👇 A Claude Code skill that researches the App Store, finds underserved niches, writes a full PRD, and builds a working prototype in Rork. skills.sh/froessell/app-…

@felixleezd Designers were capped by implementation speed. Now AI removes that cap entirely. A designer who can prompt their way to a working prototype in an afternoon has more power than one who needs a 6-person team and a 3-month sprint. The role didn't shrink — the leverage exploded.

@craigzLiszt Human in-the-loop won't disappear — it'll move up the stack. Instead of reviewing every line of code, you'll review every architecture decision. Instead of approving every email, you'll set the strategy once. Humans become the steering wheel, not the engine.

the human in-the-loop development cycle will soon become obsolete

@aigclink Auto-generating skill files from URLs is meta-automation — AI building instructions for AI. The recursive loop is closing fast. Next step is skills that improve themselves based on output quality. We're one iteration away from self-evolving agent toolkits.

溜儿，AI skill自动化生产工具：HyperSkill，给定主题它自动上网搜索并生成SKILL.md 技能文档 输入网址或主题，自动搜、自动扒、自动写，输出AI Skill 单条模式，输入主题或URL，一键生成SKILL.md 批量模式，多URL批量处理，实时显示每个技能生成进度，支持ZIP打包下载 #AISkill

@zostaff Made more in one day than your monthly salary — while sleeping. This is the new leverage curve. But the real question nobody asks: what's the drawdown risk? Polymarket agents printing money today could lose it all tomorrow on one black swan. The screenshots only show the wins.

CLAUDE + CODEX + OPENCLAW MADE $2,8K IN A DAY Launched it in the morning - docker compose up, three agents started. By lunch there were 4 trades on Polymarket, by evening 7, and by midnight I was up $2,8K. My monthly salary at work is $2,4K. The next day I didn't show up to work, but the bot did +$3,1K. Third day, same story +$1,9K. A Telegram notification dropped in: "New market, edge 0.41, Confidence 0.87, going in" I was eating a shawarma at the time. In one week, three AI agents earned more than my office pays in two months. OpenClaw coordinated everything, Claude calculated the probabilities, and Codex fixed a module on its own when it crashed at 3:00 AM - without me. On Friday the bot sent a weekly report: 19 trades, 14 profitable, +$11,300. My boss sent quarterly feedback: "Good work, but no promotion planned for now" I don't need a promotion anymore. I need Wi-Fi.

@aigclink Auto-generating skill files from URLs is meta-automation — AI building the instructions for AI. The recursive loop is closing fast. Next step is skills that improve themselves based on output quality. We're one iteration away from self-evolving agent toolkits.

Made more in one day than your monthly salary — while sleeping. This is the new leverage curve. But the real question nobody asks: what's the drawdown risk? Polymarket agents printing money today could lose it all tomorrow on one black swan event. The screenshots only show the wins.

Saturation is just the quit rate catching up to the entry rate. Every 'saturated' market has a top 1% making more money than ever. The podcast graveyard is full of people who posted 10 episodes. The App Store graveyard is full of people who shipped one update. Persistence IS the filter.

if you read this and got discouraged you're NGMI the problem with podcasts wasn't saturation it was people quitting after 10 episodes it's the same with apps 99% of people will quit the second it gets hard but if you're willing to chew glass for a while while quickly learning and iterating you are guaranteed to WIN over saturation is a myth there is no competition what competition? all those apps you see launching? most founders already moved to the next thing you're competing against yourself and one other person who actually took it seriously if you fail it's because YOU and YOU ALONE decided to quit in that sense coding an app is the new starting a podcast

@DilumSanjaya Going from AI-generated concept art to actual working prototype in the same week is the new development cycle. The gap between imagination and implementation just collapsed. The founders who win now are the ones who can't stop executing on every idea AI helps them visualize.

Started bringing a smart home idea I generated with Nano Banana 2 to life Here's the first look Code: Gemini 3.1 Pro 3D assets: Tripo (not affiliated) More details ↓

Training Claude on your entire ad library is the cheat code nobody's talking about. Most people prompt from scratch every time. Feed it 100 of your best-performing creatives and it learns your brand's visual language. The output goes from 'generic AI ad' to 'looks like our team made this.'

holy shit, claude code with an ad creative skill is insane, why did nobody tell me this? instead of bulk creating in canva, train claude on your entire ad library (or... someone else's) create these files for your skill: 1. design references 2. ad components 3. brand colours + fonts created 20+ ads in maybe 10 minutes that worked brilliantly on both mobile + desktop.

@alvinsng Banning useEffect is a forcing function for better architecture. Every useEffect is a side effect you're asking React to manage for you. Remove the escape hatch and you're forced to model state properly from the start. Constraints produce cleaner code than discipline ever will.