Shav

How to protect yourself: — npm install --ignore-scripts on any unknown repo — Run interview code in a VM or spare machine, never your main computer — Disconnect from internet before running anything unfamiliar — Use an outbound firewall like Little Snitch or GlassWire

My dear front-end developers (and anyone who’s interested in the future of interfaces): I have crawled through depths of hell to bring you, for the foreseeable years, one of the more important foundational pieces of UI engineering (if not in implementation then certainly at least in concept): Fast, accurate and comprehensive userland text measurement algorithm in pure TypeScript, usable for laying out entire web pages without CSS, bypassing DOM measurements and reflow

On maximizing Claude Code. Early on, I spent time building reusable workflows/patterns. Tedious to build, but this had a wild compounding effect as models and agent harnesses improved. I think these effects will continue to compound exponentially. No point in starting fresh every time you start a new project/feature. Invest time building subagents, skills, commands, planning, MCP tools, context engineering patterns,... Believe me when I tell you that they make a huge difference in how effective and productive you can be with Claude Code. The best part is that all these workflows are transferable to other agents like Codex. These days, I am mostly focusing on orchestration and automating code review. The workflows help a lot already, but I sometimes feel constrained by the current tools and environment. I like sandboxes and think they are going to enable a lot of the next wave of functionalities with these coding agents. I also think we are just touching the surface of context engineering and orchestrating agents. Git worktrees are great, but I am sure we can do better for scaling the work with coding agents. It's exciting to see all the progress on long-context understanding and multimodality. Both are critical to unlock even more insane capabilities with these coding agents. The former is important to enable uninterrupted longer sessions, and the latter unlocks all sorts of visual cues and grounding that will help to build more creative and unique experiences. Below is a small snapshot of a few skills I have optimized and regularly use in Claude Code. All of these leverage different models and capabilities, which will only improve as capabilities are extended and models get better. Excited for 2026 and hope to keep sharing more of my journey and keep learning from this awesome community.

🚨 RAG is broken and nobody's talking about it. Stanford just exposed the fatal flaw killing every "AI that reads your docs" product. It's called "Semantic Collapse", and it happens the moment your knowledge base hits critical mass. Here's the brutal math (and why your RAG system is already dying):

Introducing /agent by Firecrawl 🪄 Just describe what you need - with or without a URL then /agent searches, navigates, and gathers information from the widest range of websites, reaching data no other API can. Try out the research preview today.

🚨BREAKING: Claude for Excel is here. You can now use Claude directly inside Microsoft Excel: - Reads & modifies spreadsheets - Explains every change - Builds models from scratch - Fixes broken formulas Beta access now live for Max, Teams & Enterprise users.

@FlyaKiet But what does it teach you about B2B SaaS

least american opener just dropped

Introducing Claude Sonnet 4.5—the best coding model in the world. It's the strongest model for building complex agents. It's the best model at using computers. And it shows substantial gains on tests of reasoning and math.

IBM on the challengers side and Windsurf in the “leaders” quadrant. What a mess.

@ashpreetbedi 👆 The minute you do this and treat your LLM as a pure stateless function is when youre actually able to make your AI app good

Our Agents deserve a database In a world where we’ve accepted that AI products should send every trace to a 3rd party, or hand over our data to the Responses API, let me present an alternative architecture: one where we own our data. Let's give our agents a database and store each run {input=[messages], output=response} in a table. This single step unlocks: 1. Full control over context. The Agent reads the previous messages before running (and adds them to the LLM messages). I know the Responses API provides a previous_response_id that we can offload this to, but we still need to store this response_id in our database somewhere, why not just store the run and get full privacy and control? 2. Smarter context management. We can summarize, compress, enrich, and prune the conversation history. Meaning, we are in full control of the context. This is what successful Agent engineering is: removing unnecessary context, compressing and enriching it to deliver the perfect response to our user. 3. Zero dependency on 3rd party systems. No more paying egress and rention fees while begging for access. Just query your own DB. Build a quick Streamlit app, or plug into the AgentOS control plane. Your data. Your choice. Evaluation datasets, few-shot examples, multi-turn simulations, all in our control. And because we use 1 table for each agent’s sessions, our friends building LLMOps products don’t need to invent new types of datastores. We can extract user memories, implement self-learning and flag low-quality responses for review, automatically. What a beautiful world, where we ask: is AI engineering just… software engineering?

Announcing Gemini 2.5 Flash Image! (Nano Banana) from @GoogleDeepMind. This model brings you state of the art image generation! Here's 7m of demos. Today marks a leap forward in AI-powered image creativity with the launch of stronger native image generation and editing in Gemini 2.5 Flash. Here's what's new: 👤 Consistent character design: Maintain a character's identity across multiple images, changing their outfits, poses, and scenes without losing their core look. 🎨 Creative composition: Seamlessly merge elements from up to three different images into a single, cohesive masterpiece. The potential for surrealist art and unique compositions is endless. ✍️ Targeted edits with natural language: Go beyond the prompt. You can now make highly specific edits like "change the background," "restore this faded photo," or "edit the character's outfit" with simple commands. ✨ Design & identity adaptation: Apply a specific style, texture, or pattern from one image to another, unlocking new possibilities for fashion, product design, and architectural visualization. I've put together a ~7-minute video demo to walk you through these features and show just how powerful they are in action. We can't wait to see what you all create! Try the model out in Google AI Studio, Gemini app or the Gemini API.

@leerob ive had success getting cursor to write feature specific docs - works great as context when working on them again. Getting Cursor to update the rules is also handy

Becoming a better writer also makes you better with AI. I'm seeing this with devs using agents. Some are having a great time, others not so much. The difference is usually how well they write (e.g. give context). Try documenting how you build software in AGENTS​.md. It's hard! Every token matters, so being concise is important. But you also have to find the right balance of determinism vs. style and preference. You can still use type checking, linters, formatters, and other tools. Your job is steering the agent in the right direction and making clear which tools it has available (e.g. commands to run tests or linting). But you likely have additional context about how you (or your team) prefer to build software that isn’t captured by those tools. For example, more on the system design and architecture. A classic React example is: do you prefer a single component per file, or multiple components in the same file? What level of abstraction do you want? You might tell the agent to only move things to a `lib/` folder if they have been used 3 times, otherwise just copy/paste. Anyone doing this well? I'm on the hunt for AGENTS/CLAUDE/Cursor rules for inspiration. They should be concise, personalized, and very intentionally written.

The real startup hack for grit? pick a co-founder with a dog. you’ll thank me later 🐕

August at Google DeepMind be like 🚢🚢🚢 - Genie 3 - Imagen 4 Fast - Gemma 3 270M - Veo 3 Fast - Gemini Embedding - Kaggle Game Arena - Perch 2 - AI Studio <> GitHub integration and much more!

we're launching on @ProductHunt next monday! 🚀 we finally solved the design-to-code nightmare by making it ultra easy for designers to visually craft production-ready react code. who's joining us for launch day? drop a 🙋 if you're in! (cool points and more await)

I'm joining Cursor to teach the future of coding! There are millions of developers learning how to use AI and they need pragmatic advice: 1. We need to teach new developers strong foundations, so they know what to learn, and how to solve issues when debugging. 2. We need to teach experienced developers how AI can automate the tedious parts of coding, or save them time reading docs and fixing bugs. 3. We need to help developers become even more competent. AI may end up writing most of your code, but you have to review, understand, and maintain that software. This is why some experienced devs are having a great time with AI. They can ask for a pattern like "add an exponential backoff" instead of “make it more robust to errors” which may or may not work. I want to help developers become an order of magnitude more productive, and help more people contribute to building software. This is going to take a *lot* of education and retraining. So expect more videos soon, and if you have ideas for what I should teach, let me know!

Cursor for Designers Here is an open-source alternative to Lovable, v0, Figma Make, Webflow, etc.

Well which one is it?

Announcing LangChain Labs’ newest product: Open SWE Open SWE is an open-source, cloud-based asynchronous coding agent. We've been using Open SWE internally with great success. It's one of the top contributions in our open source repos, and its own repo! Try it out today, just provide an Anthropic API key. Or, you can fork the repo and deploy it yourself!

If you're a startup, your job is to punt the need for a microfrontend architecture for as long as possible. But eventually the day will come. And @vercel will be here for you 😁 We've been dogfooding this tech for a while, co-designing it with some great enterprises, and we're excited to share more broadly ↓