Nadav Lebovitch

הסוכנים שלכם הם קבלני משנה עם אמנזיה שאתם משלמים להם מחיר מלא. אז MemPalace רץ לוקאלית כשרת פקודות, מגיע עם 19 כלים, ומחזיר 96.6% מהזיכרון ברמת דיוק גבוהה, ואלגוריתם דחיסה שמכניס חודש של context ל-120 יחידות טקסט. זהו, זיכרון הוא תשתית. 🧠 github.com/milla-jovovich…

ה-IDE שייך לאתמול. Steve Yegge ו-Gene Kim צודקים – המהפכה היא לא ב-Velocity, אלא במעבר מ"איך" (קוד) ל"למה" (orchestration). התפקיד שלכם הוא לא לכתוב טקסט, אלא לבנות גבולות גזרה ל-Agent. מי שנשארו ב-IDE, נשארו מאחור 🙄 youtube.com/watch?v=7Dtu2b… #AI #VibeCoding #SoftwareEngineering

מודלים משתפרים בטירוף, ו-Skills גנריים? תכל'ס, הם מתיישנים ומבלבלים את המודל ממש מהר. הכוח האמיתי כאן הוא לבנות לעצמכם Skills משלכם, שמותאמים בול לעבודה שלכם - וכאן, חשוב לעשות את זה נכון. resources.anthropic.com/hubfs/The-Comp…

ה-CEO של Linear: הבעיה עם AI זה לא המודל, זה הממשק. 6 עקרונות לעיצוב אינטראקציית Human-Agent: • ניהול ציפיות ברור • קלט מובנה במקום רק צ'אט פתוח • גבולות שליטה מוגדרים • שקיפות התהליך • נקודות יציאה • אחריות ברורה every.to/thesis/how-to-…

@Shahriar661731 Claude

In 2026, Claude is my operating system. After 300+ hrs, I built the full blueprint: → Claude Projects + Code + Cowork → n8n MCP + SEO MCPs → Opus 4.6 + Skills Blueprint Agencies charge $5K–$10K for this. Free for you. If you want it: Like + comment “Claude” I’ll DM you.

One of the best useful and polished pieces of software I've ran into lately. Super recommended. Been waiting for Wispr Flow on Android? Win stuff on this 🎰 waitlist: wisprflow.ai/waitlist?NADAV…

@v_computer Yeah! Would love that

Introducing the first Full-Stack Vibe Coding Platform powered by Claude Code. Not only can you build a professional mobile app that accepts payments and ship it to the app store... As of today you can build a web app and deploy it to the internet with one click. In celebration of 5000 vibe coded apps published to the app store, we're giving away a free month of vibe coding for those who reply to this post. Reply and we'll DM you credits 👇

@recap_david Agent 🔥

I built an AI marketing agent to run my $100K media company. After 4 months of prompting, tooling, and integrations, I built this agent that effectively replaced my content team. Here’s how it works under the hood: → Scrapes Reddit, Hacker News, X, and Google News → Publishes a Morning Brew–style daily AI newsletter (10k daily readers) → Repurposes that content into:  • viral Twitter threads (like this one)  • short-form videos for TikTok & Instagram  • Reddit posts  • high-engagement LinkedIn updates → Produces content that’s driven millions of impressions → Generates custom, brand-aligned images for every unique asset All automated. The entire system runs through Jarvis-like voice commands, powered by ElevenLabs + n8n (see video below, I literally trained it on Jarvis from Iron Man). No manual content creation. No team to manage. Runs while I sleep. I'm no longer focusing on this business, so I'm giving all this away for free. If you want the complete system, prompting, n8n templates, & setup walkthrough: Comment “AGENT” Like & Retweet Follow me (so I can DM you) I’ll send everything over.

@kentcdodds Yet cursor published codex 5.2 working for full week uninterrupted building a browser 🤔

I like Opus 4.5 quite a bit more than GPT 5.2 Codex. With GPT 5.2 I feel like it's always stopping to check in with me on stuff even when we have a plan in place and it's irritating. Opus 4.5 does not have this issue.

@ClaudeCodeLog OMG you need Claude Code just to summerize and understand the Claude Code change log.

Chat, we are so back 🚀🚀🚀 for real this time lol Anthropic just released Claude Code 2.1.1 109 CLI, 11 flag, and 10 prompt changes, details below.

@housecor By sub agents you mean working with Claude right? I'm not familiar with such a feature in cursor.

Problem: I want to run multiple agents at once. 4 solutions: 1. Clone the repo twice. Use 2 editor instances. 2. Use Git worktree to checkout multiple branches at once. 3.Assign a GitHub issue to your agent. It runs in the cloud while you're working with a separate agent in your local IDE. 4. Use subagents. Put something like this in your instructions file: “When asked to implement a feature or a fix, always dispatch the task to the senior-developer subagent unless instructed not to do so. In the subagent context, tell the subagent to go directly into implementation. Do not nest delegation.” I tend to prefer 1 and 3 since they're simple. Other ideas? How do you handle this?

@addyosmani @googlechrome I hear you left us with a promise of vertical tabs (in canary). That's a great legacy!

After ~14 years working on Chrome, it's time for a new chapter. When I joined the @googlechrome team, I walked into a group that believed the web could be fast, simple, safe, and a joy to build on. I still believe that today, and I am continually in awe of what the team and the broader web community do every day. My last day with Chrome will be today. After that, I am excited to be moving into a new role at Google. More on that soon. Over the years I have been lucky enough to work with many teams across Chrome on new APIs, performance, DevTools, benchmarks, extensions, automation, and the broader evolution of the web platform. From Core Web Vitals and Lighthouse, to @ChromeDevTools and CDP, to efforts like Speedometer, Interop, Baseline, PWAs, AI, and our collaborations with frameworks and platforms, the constant has been the same: make the web faster, more reliable, and easier to build on. What I will remember most is not any single launch, but the way this team builds. Listening to users and developers. Measuring. Sweating the details. Making hard calls with empathy. Taking the long view while still shipping with urgency. That culture is the reason Chrome has had the impact it has on the ecosystem. To everyone in the @ChromiumDev and web community I have had the chance to work with: thank you. Thank you for the trust, the feedback, the late nights, the careful reviews, and the countless ways you pushed the platform and its tooling forward. It has been a privilege to learn from you and to build alongside you. I will still be cheering for Chrome and for the open web. If we have worked together or share an interest in the future of the web, AI, and developer experience, I would love to stay in touch as this next chapter begins.

@MessyVirgoM @DeryaTR_ Sounds awesome. So BMAD is actually everything they say it is?

@NadavLeb @DeryaTR_ Absolutely. I use almost always auto-mode or composer 1, but together with user stories & context files created using BMAD method. This breaks tasks down to small pieces where you don't need an expensive frontier model at all anymore. I'm fast, cheap and quality is just fine.

For the first time, I am fully subscribed to all 4 frontier AI models simultaneously and use them in tandem. Each model has a distinct personality, strength or perspective profile so instead of relying on a single system, I treat them as a kind of grand council of AI advisors.☺️

@MessyVirgoM @DeryaTR_ Cursor adds their own prompts on their servers before it gets to the model, so it's a bit different. Plus very expensive for most frontier models. For coding, their Composer-1 is 🔥

@DeryaTR_ Why not use Cursor? One subscription and all models there...

@yanaiEdri @NegevDev תודה ינאי! היה כיף 🙌

תודה לכל מי שהגיע היום למיטאפ של @NegevDev , תודה ל @NadavLeb על הרצאה שונה, מעשירה - ועם המון השראה!!! עד למיטאפ הבא...

story behind "why netflix built fast.com" is brilliant. so, netflix had a massive fight with ISPs around 2014-2016. ISPs were slowing down netflix on purpose. they wanted more money from netflix customers got bad streaming. but ISPs just blamed netflix. netflix had to pay comcast, verizon, at&t and time warner for direct connections to their networks. but in 2016, they launched fast dot com, clever part - It's not testing your general internet speed. It's testing your speed to netflix's servers specifically. so when someone complained about buffering, netflix could say "run fast dot com." If it's slow, the ISP is the bottleneck. suddenly millions of people had a tool to prove their ISP was the problem ISPs couldn't hide anymore. netflix positioned themselves as the transparent good guys fighting for customers while ISPs looked like greedy monopolies they solved a pr problem and a customer service problem with one simple website I guess, that's how you win a corporate war

@wesbos It's bug 2026, a distant cousin of bug 2000.

What the HECK is going on with tech? In the last week: Multiple cloud outages, x DMs totally broken, antigravity doesn't work, my watch is showing me 15 year old cal events, mac OS is a mess, email is spammed to hell and every nerd on here is talking like new AI is the second coming

@TreeS_sw @carlvellotti Well I'm selling AI workshops for tech companies for $3500. And they go out like hot buns.

@carlvellotti A course of claude code for 150$? When you can literally ask to claude code for a guide on how to use it properlly and it teaches you? I'll do it for free for anyone, DM if you want to

Every PM should be using Claude Code. So I built a HUGE course for you to learn Claude Code... IN Claude Code! 🔹 Complete guide 🔹 Make PRDs, analyze data, create decks Soon, I'll sell it for $149. For the next 24h: FREE! Follow + RT + comment "CC" & I'll DM it.

@housecor Yep. Route Context is the new client side management for react which isn't even a state management lib.

Love this feature. In short, route context can be accessed from any child routes, and you get autocomplete and type safety when you do.