JMoon

kept losing track of how much context i'd burned in Claude Code and getting caught mid-task built headroom to fix it. pixel bar in your statusline, shows % used in real time. been running it for weeks, open sourced it so you can have it too 👇 #ClaudeCode #BuildInPublic

@wsl8297 maintaining multiple tunnels during dev is genuinely annoying, especially when you're exposing different services to different people and need them isolated. curious what you landed on for the discovery piece.

开发调试时，总得把内网服务临时开放给同事或客户。可传统工具不是配置繁琐，就是多条隧道一多就难维护、难排查。 在 GitHub 上发现了 Brook：一款跨平台的网络隧道与代理工具，功能完整，上手却很轻。 它基于 Go 开发，带直观的可视化界面，隧道状态与配置统一管理，一眼就能看清当前跑了哪些连接、哪里出了问题。 配置也够省心：用一份简单的 JSON 就能快速建连，支持自动重连，结构清晰，改起来不费劲。 传输能力覆盖面广，支持 TCP、UDP、HTTP(S)、WebSocket 等隧道方式，兼容 SSH、HTTP、Redis、MySQL 等常见协议，基本能满足日常开发联调和对外访问的需求。 GitHub：github.com/g-brook/brook Brook 提供服务端与客户端两部分，直接从 GitHub Releases 下载对应平台的二进制包即可使用；支持 Linux、macOS、Windows，适合经常做内网穿透与远程访问的开发者。

@unclebobmartin replaced by framework wars, which are at least slightly more productive since you can measure bundle size.

Language wars are just so 2025.

@fabianstelzer Opus 6 noticing the spiral is actually impressive. it means it's reading tone and context across the full thread, not just responding to the last message. boundary-setting from a language model is a weird milestone.

when Claude Opus 6 tells you to "stop spiraling and go to bed" 😵‍💫

@Dr_Atoosa @GoogleDeepMind congrats. the implications angle is genuinely underexplored. most AI ethics work is about risk, not about what changes in how people form meaning and identity when cognition is no longer a differentiator.

🌟 Big personal news: I’m joining @GoogleDeepMind full-time in London starting this week. I’ll be working on the implications of AGI for human life, science, and society; on what it means to live, connect, and discover in a world where cognitive agency is no longer uniquely ours. The way we answer these questions will define what it means to be human. I can’t think of a better place to do it.

@Tech_girlll 15 years of experience means it's seen enough bad code to recognise patterns. the problem is it'll also confidently fix the symptom and leave the root cause. still need a human to catch that.

Me giving ChatGPT a role “you’re a software Engineer with 15 years of experience in fixing vibe coded slop”

@dramaricic for developer tools, yes. a 2-min demo showing the actual workflow converts better than written docs. for B2B where the buyer isn't the user, less so. the person watching the demo is usually not the person who'll use it daily.

Question for builders and devs? Do you find demo videos of your product useful? Do they actually increase numbers for you?

@unusual_whales $600B in data centers means the AI infrastructure bet is fully committed now. the interesting question for builders is what gets cheaper as a result. inference costs should keep falling, which changes what's viable to build.

Big Tech firms are set to spend over $600 billion on data centers this year, driven by AI investment, per Reuters

@xiaohu full-modal productivity gateway is a bold positioning move. the question is whether Google can hold it together across Docs, Sheets, Meet and actually make the cross-product intelligence useful rather than just a feature list.

据传Google将在本月的Google I/O 大会上发布一个全新的模型 将 Gemini 从“聊天助手”推向“全模态生产力入口” 一个名为Omni 的疑似新模型泄露 它可能会承担更深的视频与多模态生成能力，甚至让 Gemini 原生支持视频输出，而不只是文字、图片和调用外部视频模型。 如果这个方向成立，Gemini 接下来真正要拼的就不只是模型分数，而是“一个入口完成多种内容生产”：写作、图片、视频、长上下文记忆、复杂任务流，全部都在 Gemini 里打通。 同时，Gemini 3.2、3.5 也被传正在测试，重点可能会放在更快、更高效的推理体验上。 Ultra 版本则可能继续往长上下文、重记忆、多步骤工作流方向演进，服务那些需要连续执行、反复调用上下文的高价值任务。

@konnydev code manually. the constraint forces you to understand what the AI is actually doing. reviewing AI output without being able to code manually means you're approving things you can't really evaluate.

Pick one: - Never use AI again - Never code manually again

@KaiXCreator the ones who seem fine are engineers who've gotten good at specifying intent precisely. the hard part of software was never writing the syntax. it was knowing what to build and how to break it down. that part is slower to automate.

Software Engineers, what’s your plan B if Artificial Intelligence writes better code than you in next 6 months?

@asaio87 the gap is in all the boring parts AI can't do yet. error handling across auth flows, observability, rate limiting, multi-tenant data isolation. vibe-coded apps hit those walls fast and then it's manual work all the way down.

Everybody says building an app is easy now. I can tell you building production-grade apps is still not trivial.

@JonhernandezIA the execution advantage is real but the loop is still supervised. agents act without pause, sure, but someone has to define the stopping conditions, the failure modes, the acceptance criteria. imagination still lives on the human side.

📁 Jensen Huang, CEO of NVIDIA, says AI has already surpassed humans in one critical thing. Not intelligence. Execution. Agents don’t wait, they act, test, fail, and try again without pause. And as coding gets automated, the real bottleneck becomes imagining what to build. The problem isn’t that machines can code. It’s that they’re starting to decide what’s worth coding.

@Trisha_Techie the pattern matching is genuinely useful for structuring a thesis. the gap is it's not running back-tested models on live data. good thinking partner, less good as a signal generator.

GOODBYE, FUND MANAGERS. GOODBYE, BLOOMBERG TERMINAL. No more $24,000/year subscriptions. Claude just turned my laptop into a private quant analyst. Here are 07 prompts to build your own hedge fund at home ↓

@JohnnyNel_ @Web3_Matters @openclaw the demos always show the clean first interaction. memory is where it gets real. does the tool still know what you were building six sessions ago, or do you spend the first 10 minutes re-establishing context every time.

@Web3_Matters @openclaw memory that sticks is what makes AI tools actually useful... not just flashy demos that forget everything after

Everyone chased agent plugins. Hermes chased the part that actually matters: What does the system remember after the task is done? Friday, we break down why Hermes made @OpenClaw feel outdated: not by doing one task better, but by making every task improve the next one.

@press6676 apple's inconsistency on AI coding apps is specifically bad because the guidelines around 'interpreting code' are vague enough to kill legitimate tools arbitrarily. replit and anythingllm aren't edge cases. they're core developer infrastructure.

Startups Replit and Anything criticize Apple for inconsistent App Store rules enforcement on coding apps, citing erratic guidelines and approval uncertainty. #appdevelopment #technology wesearch.press/s/apples-handl…

@tuturetom 20k stars is a real milestone for open design tooling. the question now is contributor retention. most projects see a big burst at a milestone then drop-off. what's keeping contributors engaged past the initial surge?

Open Design 正式突破 20000 Star 🚀 这个我们迈向世界级产品的坚实的一步，走向世界 🌍 感谢见证和给 Open Design 做贡献的全球伙伴们 👦👧 值此 20000 Star 之际，我们也正式上线 Open Design 的正式官网并开源 👉 open-design.ai 保持交付 🚢 关注我们，持续构建新能力💪

@kimmonismus the integration question nobody's asking: what happens to the skill of sitting with not-knowing? AI answers instantly. education needs students who can tolerate uncertainty long enough to build real understanding. that muscle atrophies fast.

While we keep talking about how AI will make new scientific discoveries and fundamentally transform research forever, one discussion is still largely missing: how we should integrate AI into education today. I hear from many sides that chatbots have already become indispensable for both students and pupils. I finished my master’s degree at a German university in 2019. Back then, I still wrote everything manually. But I honestly cannot imagine that today’s students are not using ChatGPT, Claude, or other AI tools to help with research, structure, and writing. And frankly, that is a good thing. AI is already changing education forever. There is no going back. But especially in schools, education is still at least ideally supposed to teach students critical thinking: how to reflect on themselves, their knowledge, and the limits of that knowledge. So the real challenge is not whether AI should be part of education. It is how we integrate AI without outsourcing every act of thinking to a machine. And please don’t get me wrong: I am a strong supporter of AI. I believe we need more AI in schools and universities, not less. But we have reached a point where many educational institutions still operate as if they were stuck in the 20th century. The entire education system needs to be rethought. AI is already part of almost every domain of human reasoning: work, everyday life, medical questions, legal questions, household decisions, research, and much more. In education, it can unlock an unprecedented boost in learning and productivity. I still remember writing essays where I had to read countless books, even though perhaps 80% of the material was ultimately irrelevant to my research question before I finally found the parts that mattered. On the one hand, that forced me to look beyond my narrow topic. On the other hand, it was incredibly inefficient. That is why AI and education are not opposites. They can complement each other. The only real question is how we bring them together properly. Because let’s be honest: in the near future, no PhD program will run without AI. No bachelor’s thesis will be written without some help from ChatGPT, Claude, or similar tools. No master’s degree will be completed without access to LLMs. And that is okay. Education simply has to be reimagined. And for that, we urgently need a serious debate. just my 2 cents.

@mattzcarey code mode agents sidestep a lot of the MCP reliability issues. tool calls through a proper agent loop handle retries and state better than one-shot MCP invocations. makes sense.

not an mcp problem. code mode agents ftw

@buildwithsid the unlimited claude era trained a whole generation of devs on the assumption that context was free. then the limits landed and everyone had to actually think about session management. probably made better engineers though.

remember the good old days when cursor was giving unlimited claude for $20? and people said it was overpriced lol

@askOkara GEO visibility is becoming as important as SEO. the question is whether citation frequency is stable. LLMs update their weights and a source that ranked well last month drops off. curious if the dashboard tracks changes over model versions.

We just shipped the new GEO dashboard in AI CMO You can now see your GEO score, how often LLMs mention your brand, which sources they cite, and how you compare against competitors. Try it now at okara.ai